{"api":"v1","id":"chips.accelerators","title":"The accelerator record: each AI chip’s figures and dates, quoted","section":"chips","row":"figure, price, promise or shipment of one chip","page":"https://jesseahawkins.com/chips/accelerators/","fields":[{"name":"kind","type":"string","tier":null},{"name":"vendor","type":"string","tier":null},{"name":"chip","type":"string","tier":null},{"name":"field","type":"string","tier":null},{"name":"value","type":"number","tier":"quoted"},{"name":"unit","type":"string","tier":null},{"name":"printed","type":"string","tier":null},{"name":"printed_unit","type":"string","tier":null},{"name":"basis","type":"string","tier":null},{"name":"per_chips","type":"number","tier":"quoted"},{"name":"stated","type":"string","tier":null},{"name":"target","type":"string","tier":null},{"name":"window_end","type":"string","tier":null},{"name":"shipped","type":"string","tier":null},{"name":"precision","type":"string","tier":null},{"name":"status","type":"string","tier":null},{"name":"slip_months_at_most","type":"number","tier":"derived"},{"name":"quote","type":"string","tier":"quoted"},{"name":"source_url","type":"string","tier":null},{"name":"license","type":"string","tier":null},{"name":"page","type":"string","tier":null}],"tiers":["quoted","derived"],"provenance":"Each figure is the vendor’s own, quoted; the slip is this site’s count of months.","freshness":{"state":"current","asOf":"2026-11-12","oldestCheck":"2026-10-23","next":"2026-12-22","late":"2026-12-29","status":"https://jesseahawkins.com/status/","sources":["accelerators"],"notice":null},"rowCount":178,"terms":["quoted for comment"],"checked":"2026-10-23","nextCheckBy":"2026-12-22","links":{"rows":"https://jesseahawkins.com/data/v1/datasets/chips.accelerators.json","csv":"https://jesseahawkins.com/chips/accelerators.csv","releases":"https://jesseahawkins.com/data/v1/datasets/chips.accelerators/releases.json"},"latestRelease":{"id":"2026-10-23","date":"2026-10-23","sha256":"392d9bbdaa1e5aa500937d1b5766abe510f77c6210c46bdf97d564d45b435eaf","json":"https://jesseahawkins.com/data/v1/datasets/chips.accelerators/releases/2026-10-23.json","csv":"https://jesseahawkins.com/data/releases/chips/accelerators/2026-10-23.csv"},"csvSha256":"392d9bbdaa1e5aa500937d1b5766abe510f77c6210c46bdf97d564d45b435eaf","matchesLatestRelease":true,"rows":[{"kind":"figure","vendor":"NVIDIA","chip":"Tesla V100 SXM2","field":"fp16","value":125,"unit":"TFLOPS","printed":"125","printed_unit":"TFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Tesla V100 delivers 125 teraFLOPS of deep learning performance","source_url":"https://images.nvidia.com/content/technologies/volta/pdf/tesla-volta-v100-datasheet-letter-fnl-web.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-v100/"},{"kind":"figure","vendor":"NVIDIA","chip":"Tesla V100 SXM2","field":"memory","value":32,"unit":"GB","printed":"32","printed_unit":"GB","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"GPU Memory 32GB /16GB HBM2","source_url":"https://images.nvidia.com/content/technologies/volta/pdf/tesla-volta-v100-datasheet-letter-fnl-web.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-v100/"},{"kind":"figure","vendor":"NVIDIA","chip":"Tesla V100 SXM2","field":"bandwidth","value":0.9,"unit":"TB/s","printed":"900","printed_unit":"GB/s","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Memory Bandwidth 900GB/sec","source_url":"https://images.nvidia.com/content/technologies/volta/pdf/tesla-volta-v100-datasheet-letter-fnl-web.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-v100/"},{"kind":"figure","vendor":"NVIDIA","chip":"Tesla V100 SXM2","field":"power","value":300,"unit":"W","printed":"300","printed_unit":"W","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"PCIe Full Height/Length SXM2 250 W 300 W Max Power Comsumption","source_url":"https://images.nvidia.com/content/technologies/volta/pdf/tesla-volta-v100-datasheet-letter-fnl-web.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-v100/"},{"kind":"promise","vendor":"NVIDIA","chip":"Tesla V100 SXM2","field":"Volta-based DGX-1 and DGX Station orderable now, expected to ship in the third quarter","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2017-05-10","target":"2017-Q3","window_end":"2017-09-30","shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Volta-based DGX Station and DGX-1 servers are available now and expected to ship in the third quarter.","source_url":"https://nvidianews.nvidia.com/news/nvidia-advances-ai-computing-revolution-with-new-volta-based-dgx-systems","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-v100/"},{"kind":"shipment","vendor":"NVIDIA","chip":"Tesla V100 SXM2","field":"began shipping Tesla V100; shipped Volta in volume to AI customers in the quarter","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":null,"target":null,"window_end":null,"shipped":"2017-08-10","precision":"day","status":"kept","slip_months_at_most":0,"quote":"This quarter, we shipped Volta in volume to leading AI customers.","source_url":"https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-second-quarter-fiscal-2018","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-v100/"},{"kind":"figure","vendor":"NVIDIA","chip":"A100 SXM 80GB","field":"bf16","value":312,"unit":"TFLOPS","printed":"312","printed_unit":"TFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"unrecorded","slip_months_at_most":null,"quote":"BFLOAT16 Tensor Core 312 TFLOPS | 624 TFLOPS*","source_url":"https://www.nvidia.com/en-us/data-center/a100/","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-a100/"},{"kind":"figure","vendor":"NVIDIA","chip":"A100 SXM 80GB","field":"int8","value":624,"unit":"TOPS","printed":"624","printed_unit":"TOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"unrecorded","slip_months_at_most":null,"quote":"INT8 Tensor Core 624 TOPS | 1248 TOPS*","source_url":"https://www.nvidia.com/en-us/data-center/a100/","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-a100/"},{"kind":"figure","vendor":"NVIDIA","chip":"A100 SXM 80GB","field":"memory","value":80,"unit":"GB","printed":"80","printed_unit":"GB","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"unrecorded","slip_months_at_most":null,"quote":"GPU Memory 80GB HBM2e 80GB HBM2e","source_url":"https://www.nvidia.com/en-us/data-center/a100/","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-a100/"},{"kind":"figure","vendor":"NVIDIA","chip":"A100 SXM 80GB","field":"bandwidth","value":2.039,"unit":"TB/s","printed":"2,039","printed_unit":"GB/s","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"unrecorded","slip_months_at_most":null,"quote":"GPU Memory Bandwidth 1,935 GB/s 2,039 GB/s","source_url":"https://www.nvidia.com/en-us/data-center/a100/","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-a100/"},{"kind":"figure","vendor":"NVIDIA","chip":"A100 SXM 80GB","field":"power","value":400,"unit":"W","printed":"400","printed_unit":"W","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"unrecorded","slip_months_at_most":null,"quote":"Max Thermal Design Power (TDP) 300W 400W","source_url":"https://www.nvidia.com/en-us/data-center/a100/","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-a100/"},{"kind":"promise","vendor":"NVIDIA","chip":"A100 SXM 80GB","field":"A100 80GB in DGX A100 and DGX Station A100, expected to ship this quarter","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2020-11-16","target":"2020-Q4","window_end":"2020-12-31","shipped":null,"precision":null,"status":"unrecorded","slip_months_at_most":null,"quote":"The NVIDIA A100 80GB GPU is available in NVIDIA DGXTM A100 and NVIDIA DGX StationTM A100 systems, also announced today and expected to ship this quarter.","source_url":"https://nvidianews.nvidia.com/news/nvidia-doubles-down-announces-a100-80gb-gpu-supercharging-worlds-most-powerful-gpu-for-ai-supercomputing","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-a100/"},{"kind":"promise","vendor":"NVIDIA","chip":"A100 SXM 80GB","field":"partner HGX A100 systems with A100 80GB","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2020-11-16","target":"2021-H1","window_end":"2021-06-30","shipped":null,"precision":null,"status":"unrecorded","slip_months_at_most":null,"quote":"The NVIDIA A100 80GB GPU is available in NVIDIA DGXTM A100 and NVIDIA DGX StationTM A100 systems, also announced today and expected to ship this quarter.","source_url":"https://nvidianews.nvidia.com/news/nvidia-doubles-down-announces-a100-80gb-gpu-supercharging-worlds-most-powerful-gpu-for-ai-supercomputing","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-a100/"},{"kind":"figure","vendor":"NVIDIA","chip":"H100 SXM","field":"bf16","value":989.5,"unit":"TFLOPS","printed":"1,979","printed_unit":"TFLOPS","basis":"sparse","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"BFLOAT16 Tensor Core* 1,979 teraFLOPS","source_url":"https://www.nvidia.com/en-us/data-center/h100/","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-h100-sxm/"},{"kind":"figure","vendor":"NVIDIA","chip":"H100 SXM","field":"fp8","value":1979,"unit":"TFLOPS","printed":"3,958","printed_unit":"TFLOPS","basis":"sparse","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"FP8 Tensor Core* 3,958 teraFLOPS","source_url":"https://www.nvidia.com/en-us/data-center/h100/","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-h100-sxm/"},{"kind":"figure","vendor":"NVIDIA","chip":"H100 SXM","field":"memory","value":80,"unit":"GB","printed":"80","printed_unit":"GB","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"GPU Memory 80GB 94GB","source_url":"https://www.nvidia.com/en-us/data-center/h100/","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-h100-sxm/"},{"kind":"figure","vendor":"NVIDIA","chip":"H100 SXM","field":"bandwidth","value":3.35,"unit":"TB/s","printed":"3.35","printed_unit":"TB/s","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"GPU Memory Bandwidth 3.35TB/s","source_url":"https://www.nvidia.com/en-us/data-center/h100/","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-h100-sxm/"},{"kind":"figure","vendor":"NVIDIA","chip":"H100 SXM","field":"power","value":700,"unit":"W","printed":"700","printed_unit":"W","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Max Thermal Design Power (TDP) Up to 700W (configurable)","source_url":"https://www.nvidia.com/en-us/data-center/h100/","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-h100-sxm/"},{"kind":"promise","vendor":"NVIDIA","chip":"H100 SXM","field":"H100 available starting in the third quarter","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2022-03-22","target":"2022-Q3","window_end":"2022-09-30","shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"NVIDIA H100 will be available starting in the third quarter.","source_url":"https://nvidianews.nvidia.com/news/nvidia-announces-hopper-architecture-the-next-generation-of-accelerated-computing","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-h100-sxm/"},{"kind":"shipment","vendor":"NVIDIA","chip":"H100 SXM","field":"H100 in full production; partner products from October","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":null,"target":null,"window_end":null,"shipped":"2022-09-20","precision":"day","status":"kept","slip_months_at_most":0,"quote":"NVIDIA today announced that the NVIDIA H100 Tensor Core GPU is in full production","source_url":"https://nvidianews.nvidia.com/news/nvidia-hopper-in-full-production","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-h100-sxm/"},{"kind":"figure","vendor":"NVIDIA","chip":"H200 SXM","field":"bf16","value":989.5,"unit":"TFLOPS","printed":"1,979","printed_unit":"TFLOPS","basis":"sparse","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":2,"quote":"BFLOAT16 Tensor Core2 1,979 TFLOPS","source_url":"https://www.nvidia.com/en-us/data-center/h200/","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-h200/"},{"kind":"figure","vendor":"NVIDIA","chip":"H200 SXM","field":"fp8","value":1979,"unit":"TFLOPS","printed":"3,958","printed_unit":"TFLOPS","basis":"sparse","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":2,"quote":"FP8 Tensor Core2 3,958 TFLOPS","source_url":"https://www.nvidia.com/en-us/data-center/h200/","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-h200/"},{"kind":"figure","vendor":"NVIDIA","chip":"H200 SXM","field":"memory","value":141,"unit":"GB","printed":"141","printed_unit":"GB","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":2,"quote":"GPU Memory 141GB 141GB","source_url":"https://www.nvidia.com/en-us/data-center/h200/","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-h200/"},{"kind":"figure","vendor":"NVIDIA","chip":"H200 SXM","field":"bandwidth","value":4.8,"unit":"TB/s","printed":"4.8","printed_unit":"TB/s","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":2,"quote":"GPU Memory Bandwidth 4.8TB/s 4.8TB/s","source_url":"https://www.nvidia.com/en-us/data-center/h200/","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-h200/"},{"kind":"figure","vendor":"NVIDIA","chip":"H200 SXM","field":"power","value":700,"unit":"W","printed":"700","printed_unit":"W","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":2,"quote":"Max Thermal Design Power (TDP) Up to 700W (configurable)","source_url":"https://www.nvidia.com/en-us/data-center/h200/","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-h200/"},{"kind":"promise","vendor":"NVIDIA","chip":"H200 SXM","field":"H200 available from system makers and cloud providers starting in Q2 2024","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2023-11-13","target":"2024-Q2","window_end":"2024-06-30","shipped":null,"precision":null,"status":"late","slip_months_at_most":2,"quote":"The NVIDIA H200 will be available from global system manufacturers and cloud service providers starting in the second quarter of 2024.","source_url":"https://nvidianews.nvidia.com/news/nvidia-supercharges-hopper-the-worlds-leading-ai-computing-platform","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-h200/"},{"kind":"shipment","vendor":"NVIDIA","chip":"H200 SXM","field":"H200 systems generally available on CoreWeave","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":null,"target":null,"window_end":null,"shipped":"2024-08-28","precision":"day","status":"late","slip_months_at_most":2,"quote":"Revealed that H200 GPU-powered systems are now available on CoreWeave, the first cloud service provider to announce general availability.","source_url":"https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-second-quarter-fiscal-2025","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-h200/"},{"kind":"figure","vendor":"NVIDIA","chip":"GH200 Grace Hopper Superchip","field":"bf16","value":990,"unit":"TFLOPS","printed":"990","printed_unit":"TFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"BFLOAT16 Tensor Core 1,979 teraFLOPS* | 990 teraFLOPS","source_url":"https://dam-cdn.nvd.orangelogic.com/AssetLink/tkno03fa645gb7f33w60rf200d330420.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-gh200/"},{"kind":"figure","vendor":"NVIDIA","chip":"GH200 Grace Hopper Superchip","field":"fp8","value":1979,"unit":"TFLOPS","printed":"1,979","printed_unit":"TFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"FP8 Tensor Core 3,958 teraFLOPS* | 1,979 teraFLOPS","source_url":"https://dam-cdn.nvd.orangelogic.com/AssetLink/tkno03fa645gb7f33w60rf200d330420.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-gh200/"},{"kind":"figure","vendor":"NVIDIA","chip":"GH200 Grace Hopper Superchip","field":"memory","value":96,"unit":"GB","printed":"96","printed_unit":"GB","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"High-bandwidth memory (HBM) size 96GB HBM3 | 144GB HBM3e","source_url":"https://dam-cdn.nvd.orangelogic.com/AssetLink/tkno03fa645gb7f33w60rf200d330420.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-gh200/"},{"kind":"figure","vendor":"NVIDIA","chip":"GH200 Grace Hopper Superchip","field":"bandwidth","value":4,"unit":"TB/s","printed":"4","printed_unit":"TB/s","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Memory bandwidth Up to 4TB/s | Up to 4.9TB/s","source_url":"https://dam-cdn.nvd.orangelogic.com/AssetLink/tkno03fa645gb7f33w60rf200d330420.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-gh200/"},{"kind":"figure","vendor":"NVIDIA","chip":"GH200 Grace Hopper Superchip","field":"power","value":1000,"unit":"W","printed":"1000","printed_unit":"W","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Configurable 450 to 1000W (Memory + CPU + GPU)","source_url":"https://dam-cdn.nvd.orangelogic.com/AssetLink/tkno03fa645gb7f33w60rf200d330420.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-gh200/"},{"kind":"promise","vendor":"NVIDIA","chip":"GH200 Grace Hopper Superchip","field":"Grace Hopper Superchip expected to be available in the first half of 2023","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2022-03-22","target":"2023-H1","window_end":"2023-06-30","shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Both it and the Grace Hopper Superchip are expected to be available in the first half of 2023.","source_url":"https://nvidianews.nvidia.com/news/nvidia-introduces-grace-cpu-superchip","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-gh200/"},{"kind":"promise","vendor":"NVIDIA","chip":"GH200 Grace Hopper Superchip","field":"systems with GH200 expected to be available beginning later this year","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2023-05-28","target":"2023","window_end":"2023-12-31","shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Systems with GH200 Grace Hopper Superchips are expected to be available beginning later this year.","source_url":"https://nvidianews.nvidia.com/news/nvidia-grace-hopper-superchips-designed-for-accelerated-generative-ai-enter-full-production","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-gh200/"},{"kind":"shipment","vendor":"NVIDIA","chip":"GH200 Grace Hopper Superchip","field":"GH200 Grace Hopper Superchip in full production","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":null,"target":null,"window_end":null,"shipped":"2023-05-28","precision":"day","status":"kept","slip_months_at_most":0,"quote":"today announced that the NVIDIA® GH200 Grace Hopper Superchip is in full production","source_url":"https://nvidianews.nvidia.com/news/nvidia-grace-hopper-superchips-designed-for-accelerated-generative-ai-enter-full-production","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-gh200/"},{"kind":"figure","vendor":"NVIDIA","chip":"B200","field":"bf16","value":2200,"unit":"TFLOPS","printed":"2.2","printed_unit":"PFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":2,"quote":"FP16/BF16 Tensor Core Dense/Sparse 2.2/4.5 petaFLOPS 2.2/4.5 petaFLOPS","source_url":"https://dam-cdn.nvd.orangelogic.com/AssetLink/gl2l4l4812s5fw0p614s6i8bv6mi3vx5.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-b200/"},{"kind":"figure","vendor":"NVIDIA","chip":"B200","field":"fp8","value":4500,"unit":"TFLOPS","printed":"4.5","printed_unit":"PFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":2,"quote":"FP8/FP6 Tensor Core Dense/Sparse 4.5/9 petaFLOPS 4.5/9 petaFLOPS","source_url":"https://dam-cdn.nvd.orangelogic.com/AssetLink/gl2l4l4812s5fw0p614s6i8bv6mi3vx5.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-b200/"},{"kind":"figure","vendor":"NVIDIA","chip":"B200","field":"fp4","value":9000,"unit":"TFLOPS","printed":"9","printed_unit":"PFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":2,"quote":"FP4 Tensor Core Dense/Sparse 14/18 petaFLOPS 9/18 petaFLOPS","source_url":"https://dam-cdn.nvd.orangelogic.com/AssetLink/gl2l4l4812s5fw0p614s6i8bv6mi3vx5.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-b200/"},{"kind":"figure","vendor":"NVIDIA","chip":"B200","field":"memory","value":180,"unit":"GB","printed":"180","printed_unit":"GB","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":2,"quote":"186 GB HBM3E | 8 TB/s 186 GB HBM3E | 8 TB/s 180 GB HBM3E | 7.7 TB/s","source_url":"https://dam-cdn.nvd.orangelogic.com/AssetLink/y441155802qub41q118b2852i557jem5.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-b200/"},{"kind":"figure","vendor":"NVIDIA","chip":"B200","field":"bandwidth","value":7.7,"unit":"TB/s","printed":"7.7","printed_unit":"TB/s","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":2,"quote":"186 GB HBM3E | 8 TB/s 186 GB HBM3E | 8 TB/s 180 GB HBM3E | 7.7 TB/s","source_url":"https://dam-cdn.nvd.orangelogic.com/AssetLink/y441155802qub41q118b2852i557jem5.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-b200/"},{"kind":"figure","vendor":"NVIDIA","chip":"B200","field":"power","value":1000,"unit":"W","printed":"1000","printed_unit":"W","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":2,"quote":"Max thermal design power (TDP) 1100 W 1000W","source_url":"https://dam-cdn.nvd.orangelogic.com/AssetLink/gl2l4l4812s5fw0p614s6i8bv6mi3vx5.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-b200/"},{"kind":"promise","vendor":"NVIDIA","chip":"B200","field":"Blackwell-based products available from partners starting later this year","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2024-03-18","target":"2024","window_end":"2024-12-31","shipped":null,"precision":null,"status":"late","slip_months_at_most":2,"quote":"Blackwell-based products will be available from partners starting later this year.","source_url":"https://nvidianews.nvidia.com/news/nvidia-blackwell-platform-arrives-to-power-a-new-era-of-computing","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-b200/"},{"kind":"shipment","vendor":"NVIDIA","chip":"B200","field":"Blackwell in massive-scale production, billions of dollars in sales in its first quarter","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":null,"target":null,"window_end":null,"shipped":"2025-02-26","precision":"day","status":"late","slip_months_at_most":2,"quote":"We've successfully ramped up the massive-scale production of Blackwell AI supercomputers, achieving billions of dollars in sales in its first quarter.","source_url":"https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-fourth-quarter-and-fiscal-2025","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-b200/"},{"kind":"figure","vendor":"NVIDIA","chip":"GB200 Grace Blackwell Superchip","field":"bf16","value":5000,"unit":"TFLOPS","printed":"10","printed_unit":"PFLOPS","basis":"sparse","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":5,"quote":"FP16/BF16 Tensor Core2 360 PFLOPS 10 PFLOPS","source_url":"https://www.nvidia.com/en-us/data-center/gb200-nvl72/","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-gb200/"},{"kind":"figure","vendor":"NVIDIA","chip":"GB200 Grace Blackwell Superchip","field":"fp8","value":10000,"unit":"TFLOPS","printed":"20","printed_unit":"PFLOPS","basis":"sparse","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":5,"quote":"FP8/FP6 Tensor Core2 720 PFLOPS 20 PFLOPS","source_url":"https://www.nvidia.com/en-us/data-center/gb200-nvl72/","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-gb200/"},{"kind":"figure","vendor":"NVIDIA","chip":"GB200 Grace Blackwell Superchip","field":"fp4","value":20000,"unit":"TFLOPS","printed":"20","printed_unit":"PFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":5,"quote":"NVFP4 Tensor Core2 1,440 | 720 PFLOPS 40 | 20 PFLOPS","source_url":"https://www.nvidia.com/en-us/data-center/gb200-nvl72/","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-gb200/"},{"kind":"figure","vendor":"NVIDIA","chip":"GB200 Grace Blackwell Superchip","field":"memory","value":372,"unit":"GB","printed":"372","printed_unit":"GB","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":5,"quote":"GPU Memory | Bandwidth 13.4 TB HBM3E | 576 TB/s 372 GB HBM3E | 16 TB/s","source_url":"https://www.nvidia.com/en-us/data-center/gb200-nvl72/","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-gb200/"},{"kind":"figure","vendor":"NVIDIA","chip":"GB200 Grace Blackwell Superchip","field":"bandwidth","value":16,"unit":"TB/s","printed":"16","printed_unit":"TB/s","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":5,"quote":"GPU Memory | Bandwidth 13.4 TB HBM3E | 576 TB/s 372 GB HBM3E | 16 TB/s","source_url":"https://www.nvidia.com/en-us/data-center/gb200-nvl72/","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-gb200/"},{"kind":"promise","vendor":"NVIDIA","chip":"GB200 Grace Blackwell Superchip","field":"Blackwell-based products available from partners starting later this year","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2024-03-18","target":"2024","window_end":"2024-12-31","shipped":null,"precision":null,"status":"late","slip_months_at_most":5,"quote":"Blackwell-based products will be available from partners starting later this year.","source_url":"https://nvidianews.nvidia.com/news/nvidia-blackwell-platform-arrives-to-power-a-new-era-of-computing","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-gb200/"},{"kind":"shipment","vendor":"NVIDIA","chip":"GB200 Grace Blackwell Superchip","field":"Blackwell NVL72 in full-scale production across system makers and cloud providers","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":null,"target":null,"window_end":null,"shipped":"2025-05-28","precision":"day","status":"late","slip_months_at_most":5,"quote":"Our breakthrough Blackwell NVL72 AI supercomputer - a 'thinking machine' designed for reasoning- is now in full-scale production across system makers and cloud service providers","source_url":"https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-first-quarter-fiscal-2026","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-gb200/"},{"kind":"figure","vendor":"NVIDIA","chip":"B300 (Blackwell Ultra)","field":"bf16","value":2200,"unit":"TFLOPS","printed":"2.2","printed_unit":"PFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"FP16/BF16 Tensor Core Dense/Sparse 2.2/4.5 petaFLOPS 2.2/4.5 petaFLOPS","source_url":"https://dam-cdn.nvd.orangelogic.com/AssetLink/gl2l4l4812s5fw0p614s6i8bv6mi3vx5.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-b300/"},{"kind":"figure","vendor":"NVIDIA","chip":"B300 (Blackwell Ultra)","field":"fp8","value":4500,"unit":"TFLOPS","printed":"4.5","printed_unit":"PFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"FP8/FP6 Tensor Core Dense/Sparse 4.5/9 petaFLOPS 4.5/9 petaFLOPS","source_url":"https://dam-cdn.nvd.orangelogic.com/AssetLink/gl2l4l4812s5fw0p614s6i8bv6mi3vx5.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-b300/"},{"kind":"figure","vendor":"NVIDIA","chip":"B300 (Blackwell Ultra)","field":"fp4","value":14000,"unit":"TFLOPS","printed":"14","printed_unit":"PFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"FP4 Tensor Core Dense/Sparse 14/18 petaFLOPS 9/18 petaFLOPS","source_url":"https://dam-cdn.nvd.orangelogic.com/AssetLink/gl2l4l4812s5fw0p614s6i8bv6mi3vx5.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-b300/"},{"kind":"figure","vendor":"NVIDIA","chip":"B300 (Blackwell Ultra)","field":"memory","value":270,"unit":"GB","printed":"270","printed_unit":"GB","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"270 GB HBM3e | 7.7 TB/s Up to 192 GB HBM3e | 7.7 TB/s","source_url":"https://dam-cdn.nvd.orangelogic.com/AssetLink/gl2l4l4812s5fw0p614s6i8bv6mi3vx5.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-b300/"},{"kind":"figure","vendor":"NVIDIA","chip":"B300 (Blackwell Ultra)","field":"bandwidth","value":7.7,"unit":"TB/s","printed":"7.7","printed_unit":"TB/s","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"270 GB HBM3e | 7.7 TB/s Up to 192 GB HBM3e | 7.7 TB/s","source_url":"https://dam-cdn.nvd.orangelogic.com/AssetLink/gl2l4l4812s5fw0p614s6i8bv6mi3vx5.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-b300/"},{"kind":"figure","vendor":"NVIDIA","chip":"B300 (Blackwell Ultra)","field":"power","value":1100,"unit":"W","printed":"1100","printed_unit":"W","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Max thermal design power (TDP) 1100 W 1000W","source_url":"https://dam-cdn.nvd.orangelogic.com/AssetLink/gl2l4l4812s5fw0p614s6i8bv6mi3vx5.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-b300/"},{"kind":"promise","vendor":"NVIDIA","chip":"B300 (Blackwell Ultra)","field":"Blackwell Ultra-based products expected from partners from the second half of 2025","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2025-03-18","target":"2025-H2","window_end":"2025-12-31","shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Blackwell Ultra-based products are expected to be available from partners starting from the second half of 2025.","source_url":"https://nvidianews.nvidia.com/news/nvidia-blackwell-ultra-ai-factory-platform-paves-way-for-age-of-ai-reasoning","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-b300/"},{"kind":"shipment","vendor":"NVIDIA","chip":"B300 (Blackwell Ultra)","field":"Blackwell Ultra production ramping at full speed","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":null,"target":null,"window_end":null,"shipped":"2025-08-27","precision":"day","status":"kept","slip_months_at_most":0,"quote":"production of Blackwell Ultra is ramping at full speed, and demand is extraordinary","source_url":"https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-second-quarter-fiscal-2026","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-b300/"},{"kind":"figure","vendor":"NVIDIA","chip":"Rubin GPU","field":"bf16","value":4000,"unit":"TFLOPS","printed":"8","printed_unit":"PFLOPS","basis":"dense","per_chips":2,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"FP16/BF163 288 PFLOPS 8 PFLOPS 16 PFLOPS","source_url":"https://dam-cdn.nvd.orangelogic.com/AssetLink/56p68o47y6f1yucpubl0ump1c0aif422.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-rubin/"},{"kind":"figure","vendor":"NVIDIA","chip":"Rubin GPU","field":"fp8","value":17500,"unit":"TFLOPS","printed":"35","printed_unit":"PFLOPS","basis":"dense","per_chips":2,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"FP8/FP6 Training3 1,260 PFLOPS 35 PFLOPS 70 PFLOPS","source_url":"https://dam-cdn.nvd.orangelogic.com/AssetLink/56p68o47y6f1yucpubl0ump1c0aif422.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-rubin/"},{"kind":"figure","vendor":"NVIDIA","chip":"Rubin GPU","field":"fp4","value":35000,"unit":"TFLOPS","printed":"70","printed_unit":"PFLOPS","basis":"dense","per_chips":2,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"NVFP4 Training 2,520 PFLOPS 70 PFLOPS 140 PFLOPS","source_url":"https://dam-cdn.nvd.orangelogic.com/AssetLink/56p68o47y6f1yucpubl0ump1c0aif422.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-rubin/"},{"kind":"figure","vendor":"NVIDIA","chip":"Rubin GPU","field":"memory","value":288,"unit":"GB","printed":"576","printed_unit":"GB","basis":null,"per_chips":2,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"20.7 TB HBM4 | 1,400 TB/s 576 GB HBM4 | 38.5 TB/s","source_url":"https://dam-cdn.nvd.orangelogic.com/AssetLink/56p68o47y6f1yucpubl0ump1c0aif422.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-rubin/"},{"kind":"figure","vendor":"NVIDIA","chip":"Rubin GPU","field":"bandwidth","value":19.25,"unit":"TB/s","printed":"38.5","printed_unit":"TB/s","basis":null,"per_chips":2,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"20.7 TB HBM4 | 1,400 TB/s 576 GB HBM4 | 38.5 TB/s","source_url":"https://dam-cdn.nvd.orangelogic.com/AssetLink/56p68o47y6f1yucpubl0ump1c0aif422.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-rubin/"},{"kind":"promise","vendor":"NVIDIA","chip":"Rubin GPU","field":"Rubin-based products available from partners in the second half of 2026","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2026-01-05","target":"2026-H2","window_end":"2026-12-31","shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"NVIDIA Rubin is in full production, and Rubin-based products will be available from partners the second half of 2026.","source_url":"https://nvidianews.nvidia.com/news/rubin-platform-ai-supercomputer","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-rubin/"},{"kind":"promise","vendor":"NVIDIA","chip":"Rubin GPU","field":"production shipments of Vera Rubin to begin this fall","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2026-05-31","target":"2026-H2","window_end":"2026-12-31","shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Production shipments of Vera Rubin are set to begin starting this fall.","source_url":"https://nvidianews.nvidia.com/news/vera-rubin-full-production-agentic-ai-factory","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-rubin/"},{"kind":"shipment","vendor":"NVIDIA","chip":"Rubin GPU","field":"Vera Rubin ramping into full production, with racks running at partners","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":null,"target":null,"window_end":null,"shipped":"2026-08-26","precision":"day","status":"kept","slip_months_at_most":0,"quote":"Announced the NVIDIA Vera Rubin platform is ramping into full production with racks running at partners including CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure and Nebius.","source_url":"https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-second-quarter-fiscal-2027","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/nvidia-rubin/"},{"kind":"figure","vendor":"AMD","chip":"Instinct MI250X","field":"bf16","value":383,"unit":"TFLOPS","printed":"383","printed_unit":"TFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"delivered","slip_months_at_most":null,"quote":"Peak bfloat16 383 TFLOPs","source_url":"https://www.amd.com/en/products/accelerators/instinct/mi200/mi250x.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi250x/"},{"kind":"figure","vendor":"AMD","chip":"Instinct MI250X","field":"int8","value":383,"unit":"TOPS","printed":"383","printed_unit":"TOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"delivered","slip_months_at_most":null,"quote":"Peak INT8 Performance 383 TOPs","source_url":"https://www.amd.com/en/products/accelerators/instinct/mi200/mi250x.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi250x/"},{"kind":"figure","vendor":"AMD","chip":"Instinct MI250X","field":"memory","value":128,"unit":"GB","printed":"128","printed_unit":"GB","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"delivered","slip_months_at_most":null,"quote":"Dedicated Memory Size 128 GB","source_url":"https://www.amd.com/en/products/accelerators/instinct/mi200/mi250x.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi250x/"},{"kind":"figure","vendor":"AMD","chip":"Instinct MI250X","field":"bandwidth","value":3.2,"unit":"TB/s","printed":"3.2","printed_unit":"TB/s","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"delivered","slip_months_at_most":null,"quote":"Peak Memory Bandwidth 3.2 TB/s","source_url":"https://www.amd.com/en/products/accelerators/instinct/mi200/mi250x.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi250x/"},{"kind":"figure","vendor":"AMD","chip":"Instinct MI250X","field":"power","value":560,"unit":"W","printed":"560","printed_unit":"W","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"delivered","slip_months_at_most":null,"quote":"Thermal Design Power (TDP) 500W | 560W","source_url":"https://www.amd.com/en/products/accelerators/instinct/mi200/mi250x.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi250x/"},{"kind":"shipment","vendor":"AMD","chip":"Instinct MI250X","field":"available from HPE in the HPE Cray EX supercomputer","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":null,"target":null,"window_end":null,"shipped":"2021-11-08","precision":"day","status":"delivered","slip_months_at_most":null,"quote":"The AMD MI250X accelerator is currently available from HPE in the HPE Cray EX Supercomputer","source_url":"https://ir.amd.com/news-events/press-releases/detail/1032/new-amd-instinct-mi200-series-accelerators-bring-leadership-hpc-and-ai-performance-to-power-exascale-systems-and-more","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi250x/"},{"kind":"figure","vendor":"AMD","chip":"Instinct MI300X","field":"bf16","value":1300,"unit":"TFLOPS","printed":"1.3","printed_unit":"PFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Peak bfloat16 1.3 PFLOPs","source_url":"https://www.amd.com/en/products/accelerators/instinct/mi300/mi300x.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi300x/"},{"kind":"figure","vendor":"AMD","chip":"Instinct MI300X","field":"fp8","value":2610,"unit":"TFLOPS","printed":"2.61","printed_unit":"PFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Peak Eight-bit Precision (FP8) Performance (E5M2, E4M3) 2.61 PFLOPs","source_url":"https://www.amd.com/en/products/accelerators/instinct/mi300/mi300x.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi300x/"},{"kind":"figure","vendor":"AMD","chip":"Instinct MI300X","field":"memory","value":192,"unit":"GB","printed":"192","printed_unit":"GB","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Dedicated Memory Size 192 GB","source_url":"https://www.amd.com/en/products/accelerators/instinct/mi300/mi300x.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi300x/"},{"kind":"figure","vendor":"AMD","chip":"Instinct MI300X","field":"bandwidth","value":5.3,"unit":"TB/s","printed":"5.3","printed_unit":"TB/s","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Peak Memory Bandwidth 5.3 TB/s","source_url":"https://www.amd.com/en/products/accelerators/instinct/mi300/mi300x.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi300x/"},{"kind":"figure","vendor":"AMD","chip":"Instinct MI300X","field":"power","value":750,"unit":"W","printed":"750","printed_unit":"W","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Typical Board Power (TBP) 750W","source_url":"https://www.amd.com/en/products/accelerators/instinct/mi300/mi300x.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi300x/"},{"kind":"promise","vendor":"AMD","chip":"Instinct MI300X","field":"launch","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2023-06-13","target":"2023","window_end":"2023-12-31","shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"led by the launch of our Instinct MI300 accelerators planned for later this year","source_url":"https://ir.amd.com/news-events/press-releases/detail/1136/amd-expands-leadership-data-center-portfolio-with-new-epyc-cpus-and-shares-details-on-next-generation-amd-instinct-accelerator-and-software-enablement-for-generative-ai","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi300x/"},{"kind":"promise","vendor":"AMD","chip":"Instinct MI300X","field":"launch and production ramp","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2023-08-01","target":"2023-Q4","window_end":"2023-12-31","shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"on-track to launch and ramp production of MI300 accelerators in the fourth quarter","source_url":"https://www.sec.gov/Archives/edgar/data/2488/000000248823000135/q22023991.htm","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi300x/"},{"kind":"promise","vendor":"AMD","chip":"Instinct MI300X","field":"volume production","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2023-10-31","target":"2023-Q4","window_end":"2023-12-31","shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"AMD InstinctTM MI300A and MI300X GPUs are on track for volume production in the fourth quarter","source_url":"https://www.sec.gov/Archives/edgar/data/2488/000000248823000192/q32023991.htm","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi300x/"},{"kind":"shipment","vendor":"AMD","chip":"Instinct MI300X","field":"AMD announced availability","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":null,"target":null,"window_end":null,"shipped":"2023-12-06","precision":"day","status":"kept","slip_months_at_most":0,"quote":"Today, AMD (NASDAQ: AMD) announced the availability of the AMD InstinctTM MI300X accelerators","source_url":"https://ir.amd.com/news-events/press-releases/detail/1173/amd-delivers-leadership-portfolio-of-data-center-ai-solutions-with-amd-instinct-mi300-series","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi300x/"},{"kind":"figure","vendor":"AMD","chip":"Instinct MI300A","field":"bf16","value":980.6,"unit":"TFLOPS","printed":"980.6","printed_unit":"TFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Peak bfloat16 980.6 TFLOPs","source_url":"https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi300a/"},{"kind":"figure","vendor":"AMD","chip":"Instinct MI300A","field":"fp8","value":1960,"unit":"TFLOPS","printed":"1.96","printed_unit":"PFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Peak Eight-bit Precision (FP8) Performance (E5M2, E4M3) 1.96 PFLOPs","source_url":"https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi300a/"},{"kind":"figure","vendor":"AMD","chip":"Instinct MI300A","field":"memory","value":128,"unit":"GB","printed":"128","printed_unit":"GB","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Dedicated Memory Size 128 GB","source_url":"https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi300a/"},{"kind":"figure","vendor":"AMD","chip":"Instinct MI300A","field":"bandwidth","value":5.3,"unit":"TB/s","printed":"5.3","printed_unit":"TB/s","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Peak Memory Bandwidth 5.3 TB/s","source_url":"https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi300a/"},{"kind":"figure","vendor":"AMD","chip":"Instinct MI300A","field":"power","value":760,"unit":"W","printed":"760","printed_unit":"W","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Thermal Design Power (TDP) 550W | 760W","source_url":"https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi300a/"},{"kind":"promise","vendor":"AMD","chip":"Instinct MI300A","field":"first CDNA 3 products (the MI300 family)","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2022-06-09","target":"2023","window_end":"2023-12-31","shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"The first AMD CDNA 3 architecture-based products are planned for 2023","source_url":"https://ir.amd.com/news-events/press-releases/detail/1078/amd-details-strategy-to-drive-next-phase-of-growth-across-300-billion-market-for-high-performance-and-adaptive-computing-solutions","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi300a/"},{"kind":"promise","vendor":"AMD","chip":"Instinct MI300A","field":"launch","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2023-06-13","target":"2023","window_end":"2023-12-31","shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"led by the launch of our Instinct MI300 accelerators planned for later this year","source_url":"https://ir.amd.com/news-events/press-releases/detail/1136/amd-expands-leadership-data-center-portfolio-with-new-epyc-cpus-and-shares-details-on-next-generation-amd-instinct-accelerator-and-software-enablement-for-generative-ai","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi300a/"},{"kind":"promise","vendor":"AMD","chip":"Instinct MI300A","field":"launch and production ramp","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2023-08-01","target":"2023-Q4","window_end":"2023-12-31","shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"on-track to launch and ramp production of MI300 accelerators in the fourth quarter","source_url":"https://www.sec.gov/Archives/edgar/data/2488/000000248823000135/q22023991.htm","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi300a/"},{"kind":"promise","vendor":"AMD","chip":"Instinct MI300A","field":"volume production","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2023-10-31","target":"2023-Q4","window_end":"2023-12-31","shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"AMD InstinctTM MI300A and MI300X GPUs are on track for volume production in the fourth quarter","source_url":"https://www.sec.gov/Archives/edgar/data/2488/000000248823000192/q32023991.htm","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi300a/"},{"kind":"shipment","vendor":"AMD","chip":"Instinct MI300A","field":"AMD announced availability","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":null,"target":null,"window_end":null,"shipped":"2023-12-06","precision":"day","status":"kept","slip_months_at_most":0,"quote":"as well as the AMD InstinctTM MI300A accelerated processing unit (APU)","source_url":"https://ir.amd.com/news-events/press-releases/detail/1173/amd-delivers-leadership-portfolio-of-data-center-ai-solutions-with-amd-instinct-mi300-series","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi300a/"},{"kind":"figure","vendor":"AMD","chip":"Instinct MI325X","field":"bf16","value":1300,"unit":"TFLOPS","printed":"1.3","printed_unit":"PFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"unrecorded","slip_months_at_most":null,"quote":"Peak bfloat16 1.3 PFLOPs","source_url":"https://www.amd.com/en/products/accelerators/instinct/mi300/mi325x.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi325x/"},{"kind":"figure","vendor":"AMD","chip":"Instinct MI325X","field":"fp8","value":2610,"unit":"TFLOPS","printed":"2.61","printed_unit":"PFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"unrecorded","slip_months_at_most":null,"quote":"Peak Eight-bit Precision (FP8) Performance (E5M2, E4M3) 2.61 PFLOPs","source_url":"https://www.amd.com/en/products/accelerators/instinct/mi300/mi325x.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi325x/"},{"kind":"figure","vendor":"AMD","chip":"Instinct MI325X","field":"memory","value":256,"unit":"GB","printed":"256","printed_unit":"GB","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"unrecorded","slip_months_at_most":null,"quote":"Dedicated Memory Size 256 GB","source_url":"https://www.amd.com/en/products/accelerators/instinct/mi300/mi325x.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi325x/"},{"kind":"figure","vendor":"AMD","chip":"Instinct MI325X","field":"bandwidth","value":6,"unit":"TB/s","printed":"6","printed_unit":"TB/s","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"unrecorded","slip_months_at_most":null,"quote":"Peak Memory Bandwidth 6 TB/s","source_url":"https://www.amd.com/en/products/accelerators/instinct/mi300/mi325x.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi325x/"},{"kind":"figure","vendor":"AMD","chip":"Instinct MI325X","field":"power","value":1000,"unit":"W","printed":"1000","printed_unit":"W","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"unrecorded","slip_months_at_most":null,"quote":"Typical Board Power (TBP) 1000W","source_url":"https://www.amd.com/en/products/accelerators/instinct/mi300/mi325x.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi325x/"},{"kind":"promise","vendor":"AMD","chip":"Instinct MI325X","field":"availability","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2024-06-02","target":"2024-Q4","window_end":"2024-12-31","shipped":null,"precision":null,"status":"unrecorded","slip_months_at_most":null,"quote":"The expanded roadmap includes the AMD Instinct MI325X accelerators, with planned availability in Q4 2024","source_url":"https://www.amd.com/en/newsroom/press-releases/2024-6-2-amd-extends-ai-and-high-performance-leadership-in-.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi325x/"},{"kind":"promise","vendor":"AMD","chip":"Instinct MI325X","field":"availability","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2024-07-30","target":"2024-Q4","window_end":"2024-12-31","shipped":null,"precision":null,"status":"unrecorded","slip_months_at_most":null,"quote":"The roadmap includes the new AMD Instinct MI325X accelerator, planned to be available in Q4 2024","source_url":"https://www.sec.gov/Archives/edgar/data/2488/000000248824000121/q22024991.htm","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi325x/"},{"kind":"promise","vendor":"AMD","chip":"Instinct MI325X","field":"production shipments","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2024-10-10","target":"2024-Q4","window_end":"2024-12-31","shipped":null,"precision":null,"status":"unrecorded","slip_months_at_most":null,"quote":"AMD Instinct MI325X accelerators are currently on track for production shipments in Q4 2024","source_url":"https://ir.amd.com/news-events/press-releases/detail/1220/amd-delivers-leadership-ai-performance-with-amd-instinct-mi325x-accelerators","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi325x/"},{"kind":"figure","vendor":"AMD","chip":"Instinct MI355X","field":"bf16","value":2500,"unit":"TFLOPS","printed":"2.5","printed_unit":"PFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Peak bfloat16 (BF16) Matrix Performance 2.5 PFLOPs","source_url":"https://www.amd.com/en/products/accelerators/instinct/mi350/mi355x.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi355x/"},{"kind":"figure","vendor":"AMD","chip":"Instinct MI355X","field":"fp8","value":5000,"unit":"TFLOPS","printed":"5","printed_unit":"PFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Peak Open Compute Project Eight-bit Precision Matrix (OCP-FP8) Performance (E5M2, E4M3) 5 PFLOPs","source_url":"https://www.amd.com/en/products/accelerators/instinct/mi350/mi355x.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi355x/"},{"kind":"figure","vendor":"AMD","chip":"Instinct MI355X","field":"fp4","value":10100,"unit":"TFLOPS","printed":"10.1","printed_unit":"PFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Peak Microscaling Four-bit Precision Matrix (MXFP4) Performance 10.1 PFLOPs","source_url":"https://www.amd.com/en/products/accelerators/instinct/mi350/mi355x.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi355x/"},{"kind":"figure","vendor":"AMD","chip":"Instinct MI355X","field":"memory","value":288,"unit":"GB","printed":"288","printed_unit":"GB","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Dedicated Memory Size 288 GB","source_url":"https://www.amd.com/en/products/accelerators/instinct/mi350/mi355x.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi355x/"},{"kind":"figure","vendor":"AMD","chip":"Instinct MI355X","field":"bandwidth","value":8,"unit":"TB/s","printed":"8","printed_unit":"TB/s","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Peak Memory Bandwidth 8 TB/s","source_url":"https://www.amd.com/en/products/accelerators/instinct/mi350/mi355x.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi355x/"},{"kind":"figure","vendor":"AMD","chip":"Instinct MI355X","field":"power","value":1400,"unit":"W","printed":"1400","printed_unit":"W","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Typical Board Power (TBP) 1400W","source_url":"https://www.amd.com/en/products/accelerators/instinct/mi350/mi355x.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi355x/"},{"kind":"promise","vendor":"AMD","chip":"Instinct MI355X","field":"MI350 Series (CDNA 4) expected","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2024-06-02","target":"2025","window_end":"2025-12-31","shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"The next-generation AMD CDNATM 4 architecture, expected in 2025, will power the AMD Instinct MI350 Series","source_url":"https://www.amd.com/en/newsroom/press-releases/2024-6-2-amd-extends-ai-and-high-performance-leadership-in-.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi355x/"},{"kind":"promise","vendor":"AMD","chip":"Instinct MI355X","field":"MI350 Series availability","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2024-10-10","target":"2025-H2","window_end":"2025-12-31","shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"The AMD Instinct MI350 series accelerators are on track to be available during the second half of 2025","source_url":"https://ir.amd.com/news-events/press-releases/detail/1220/amd-delivers-leadership-ai-performance-with-amd-instinct-mi325x-accelerators","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi355x/"},{"kind":"shipment","vendor":"AMD","chip":"Instinct MI355X","field":"Q3 2025 data center revenue driven by MI350 Series GPU sales (the series includes the MI355X)","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":null,"target":null,"window_end":null,"shipped":"2025-07","precision":"quarter","status":"kept","slip_months_at_most":0,"quote":"Data Center segment revenue was $4.3 billion, up 22% year-over-year primarily driven by strong demand for 5th Gen AMD EPYCTM processors and AMD Instinct MI350 Series GPUs","source_url":"https://www.sec.gov/Archives/edgar/data/2488/000000248825000163/q32025991.htm","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/amd-mi355x/"},{"kind":"figure","vendor":"Google","chip":"TPU v4","field":"bf16","value":275,"unit":"TFLOPS","printed":"275","printed_unit":"TFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":10,"quote":"Peak compute per chip 275 teraflops (bf16 or int8)","source_url":"https://cloud.google.com/tpu/docs/v4","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v4/"},{"kind":"figure","vendor":"Google","chip":"TPU v4","field":"int8","value":275,"unit":"TOPS","printed":"275","printed_unit":"TOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":10,"quote":"Peak compute per chip 275 teraflops (bf16 or int8)","source_url":"https://cloud.google.com/tpu/docs/v4","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v4/"},{"kind":"figure","vendor":"Google","chip":"TPU v4","field":"memory","value":32,"unit":"GiB","printed":"32","printed_unit":"GiB","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":10,"quote":"HBM2 capacity and bandwidth 32 GiB, 1200 GBps","source_url":"https://cloud.google.com/tpu/docs/v4","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v4/"},{"kind":"figure","vendor":"Google","chip":"TPU v4","field":"bandwidth","value":1.2,"unit":"TB/s","printed":"1200","printed_unit":"GB/s","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":10,"quote":"HBM2 capacity and bandwidth 32 GiB, 1200 GBps","source_url":"https://cloud.google.com/tpu/docs/v4","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v4/"},{"kind":"price","vendor":"Google","chip":"TPU v4","field":"usd_per_chip_hour","value":3.22,"unit":"USD","printed":"$3.22","printed_unit":"USD","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":10,"quote":"TPU v4 pod us-central2 Oklahoma $3.22","source_url":"https://cloud.google.com/tpu/pricing","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v4/"},{"kind":"promise","vendor":"Google","chip":"TPU v4","field":"TPU v4 pods available to Cloud customers","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2021-05-18","target":"2021","window_end":"2021-12-31","shipped":null,"precision":null,"status":"late","slip_months_at_most":10,"quote":"They'll be available to our Cloud customers later this year.","source_url":"https://blog.google/technology/developers/io21-helpful-google/","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v4/"},{"kind":"shipment","vendor":"Google","chip":"TPU v4","field":"Cloud TPU v4 Pods generally available","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":null,"target":null,"window_end":null,"shipped":"2022-10-12","precision":"day","status":"late","slip_months_at_most":10,"quote":"Cloud TPU v4 Pods - Now in General Availability (GA)","source_url":"https://cloud.google.com/blog/products/infrastructure-modernization/open-infrastructure-announcements-at-google-cloud-next","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v4/"},{"kind":"figure","vendor":"Google","chip":"TPU v5e","field":"bf16","value":197,"unit":"TFLOPS","printed":"197","printed_unit":"TFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"delivered","slip_months_at_most":null,"quote":"Peak compute per chip (bf16) 197 TFLOPs","source_url":"https://cloud.google.com/tpu/docs/v5e","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v5e/"},{"kind":"figure","vendor":"Google","chip":"TPU v5e","field":"int8","value":393,"unit":"TOPS","printed":"393","printed_unit":"TOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"delivered","slip_months_at_most":null,"quote":"Peak compute per chip (Int8) 393 TOPs","source_url":"https://cloud.google.com/tpu/docs/v5e","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v5e/"},{"kind":"figure","vendor":"Google","chip":"TPU v5e","field":"memory","value":16,"unit":"GB","printed":"16","printed_unit":"GB","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"delivered","slip_months_at_most":null,"quote":"HBM capacity per chip 16 GB","source_url":"https://cloud.google.com/tpu/docs/v5e","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v5e/"},{"kind":"figure","vendor":"Google","chip":"TPU v5e","field":"bandwidth","value":0.859,"unit":"TB/s","printed":"800","printed_unit":"GiB/s","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"delivered","slip_months_at_most":null,"quote":"HBM bandwidth per chip 800 GiBps","source_url":"https://cloud.google.com/tpu/docs/v5e","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v5e/"},{"kind":"price","vendor":"Google","chip":"TPU v5e","field":"usd_per_chip_hour","value":1.2,"unit":"USD","printed":"$1.20","printed_unit":"USD","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"delivered","slip_months_at_most":null,"quote":"TPU v5e us-central1 Iowa $1.20","source_url":"https://cloud.google.com/tpu/pricing","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v5e/"},{"kind":"shipment","vendor":"Google","chip":"TPU v5e","field":"Cloud TPU v5e generally available","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":null,"target":null,"window_end":null,"shipped":"2023-11-09","precision":"day","status":"delivered","slip_months_at_most":null,"quote":"Second, Cloud TPU v5e is now generally available","source_url":"https://cloud.google.com/blog/products/compute/announcing-cloud-tpu-v5e-in-ga","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v5e/"},{"kind":"figure","vendor":"Google","chip":"TPU v5p","field":"bf16","value":459,"unit":"TFLOPS","printed":"459","printed_unit":"TFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"delivered","slip_months_at_most":null,"quote":"Peak compute per chip (BF16) (TFLOPs) 459","source_url":"https://cloud.google.com/tpu/docs/v5p","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v5p/"},{"kind":"figure","vendor":"Google","chip":"TPU v5p","field":"fp8","value":459,"unit":"TFLOPS","printed":"459","printed_unit":"TFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"delivered","slip_months_at_most":null,"quote":"Peak compute per chip (FP8) (TFLOPs) 459","source_url":"https://cloud.google.com/tpu/docs/v5p","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v5p/"},{"kind":"figure","vendor":"Google","chip":"TPU v5p","field":"memory","value":95,"unit":"GiB","printed":"95","printed_unit":"GiB","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"delivered","slip_months_at_most":null,"quote":"HBM capacity per chip (GiB) 95","source_url":"https://cloud.google.com/tpu/docs/v5p","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v5p/"},{"kind":"figure","vendor":"Google","chip":"TPU v5p","field":"bandwidth","value":2.765,"unit":"TB/s","printed":"2765","printed_unit":"GB/s","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"delivered","slip_months_at_most":null,"quote":"HBM bandwidth per chip (GBps) 2765","source_url":"https://cloud.google.com/tpu/docs/v5p","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v5p/"},{"kind":"price","vendor":"Google","chip":"TPU v5p","field":"usd_per_chip_hour","value":4.2,"unit":"USD","printed":"$4.20","printed_unit":"USD","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"delivered","slip_months_at_most":null,"quote":"TPU v5p us-east5 Columbus $4.20","source_url":"https://cloud.google.com/tpu/pricing","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v5p/"},{"kind":"shipment","vendor":"Google","chip":"TPU v5p","field":"Cloud TPU v5p generally available","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":null,"target":null,"window_end":null,"shipped":"2024-04-09","precision":"day","status":"delivered","slip_months_at_most":null,"quote":"We're thrilled to announce the general availability of Cloud TPU v5p","source_url":"https://cloud.google.com/blog/products/compute/whats-new-with-google-clouds-ai-hypercomputer-architecture","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v5p/"},{"kind":"figure","vendor":"Google","chip":"TPU v6e","field":"bf16","value":918,"unit":"TFLOPS","printed":"918","printed_unit":"TFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Peak compute per chip (bf16) 918 TFLOPs","source_url":"https://cloud.google.com/tpu/docs/v6e","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v6e/"},{"kind":"figure","vendor":"Google","chip":"TPU v6e","field":"fp8","value":918,"unit":"TFLOPS","printed":"918","printed_unit":"TFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Peak compute per chip (FP8) (TFLOPs) 459 918 4614","source_url":"https://cloud.google.com/tpu/docs/tpu7x","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v6e/"},{"kind":"figure","vendor":"Google","chip":"TPU v6e","field":"memory","value":32,"unit":"GB","printed":"32","printed_unit":"GB","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"HBM capacity per chip 32 GB","source_url":"https://cloud.google.com/tpu/docs/v6e","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v6e/"},{"kind":"figure","vendor":"Google","chip":"TPU v6e","field":"bandwidth","value":1.638,"unit":"TB/s","printed":"1638","printed_unit":"GB/s","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"HBM bandwidth per chip 1638 GBps","source_url":"https://cloud.google.com/tpu/docs/v6e","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v6e/"},{"kind":"price","vendor":"Google","chip":"TPU v6e","field":"usd_per_chip_hour","value":2.7,"unit":"USD","printed":"$2.70","printed_unit":"USD","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Trillium us-east1 South Carolina $2.70","source_url":"https://cloud.google.com/tpu/pricing","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v6e/"},{"kind":"promise","vendor":"Google","chip":"TPU v6e","field":"Trillium TPUs available","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2024-05-15","target":"2024","window_end":"2024-12-31","shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"The sixth-generation Trillium TPUs are a culmination of over a decade of research and innovation and will be available later this year.","source_url":"https://cloud.google.com/blog/products/compute/introducing-trillium-6th-gen-tpus","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v6e/"},{"kind":"shipment","vendor":"Google","chip":"TPU v6e","field":"Trillium generally available for Google Cloud customers","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":null,"target":null,"window_end":null,"shipped":"2024-12-11","precision":"day","status":"kept","slip_months_at_most":0,"quote":"Earlier this year, we announced Trillium, our sixth-generation and most performant TPU to date. Today, it is generally available for Google Cloud customers.","source_url":"https://cloud.google.com/blog/products/compute/trillium-tpu-is-ga","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v6e/"},{"kind":"figure","vendor":"Google","chip":"TPU7x (Ironwood)","field":"bf16","value":2307,"unit":"TFLOPS","printed":"2307","printed_unit":"TFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":3,"quote":"Peak compute per chip (BF16) (TFLOPs) 459 918 2307","source_url":"https://cloud.google.com/tpu/docs/tpu7x","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v7/"},{"kind":"figure","vendor":"Google","chip":"TPU7x (Ironwood)","field":"fp8","value":4614,"unit":"TFLOPS","printed":"4614","printed_unit":"TFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":3,"quote":"Peak compute per chip (FP8) (TFLOPs) 459 918 4614","source_url":"https://cloud.google.com/tpu/docs/tpu7x","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v7/"},{"kind":"figure","vendor":"Google","chip":"TPU7x (Ironwood)","field":"memory","value":192,"unit":"GiB","printed":"192","printed_unit":"GiB","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":3,"quote":"Each chip is equipped with 192 GiB of HBM","source_url":"https://cloud.google.com/tpu/docs/tpu7x","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v7/"},{"kind":"figure","vendor":"Google","chip":"TPU7x (Ironwood)","field":"bandwidth","value":7.38,"unit":"TB/s","printed":"7380","printed_unit":"GB/s","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":3,"quote":"HBM bandwidth per chip (GBps) 2765 1638 7380","source_url":"https://cloud.google.com/tpu/docs/tpu7x","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v7/"},{"kind":"price","vendor":"Google","chip":"TPU7x (Ironwood)","field":"usd_per_chip_hour","value":12,"unit":"USD","printed":"$12.00","printed_unit":"USD","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":3,"quote":"Ironwood us-central1 Iowa $12.00","source_url":"https://cloud.google.com/tpu/pricing","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v7/"},{"kind":"promise","vendor":"Google","chip":"TPU7x (Ironwood)","field":"Ironwood available to Google Cloud customers","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2025-04-09","target":"2025","window_end":"2025-12-31","shipped":null,"precision":null,"status":"late","slip_months_at_most":3,"quote":"when it becomes available later this year","source_url":"https://blog.google/products/google-cloud/ironwood-tpu-age-of-inference/","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v7/"},{"kind":"promise","vendor":"Google","chip":"TPU7x (Ironwood)","field":"general availability \"in the coming weeks\", read as by the end of 2025","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2025-11-06","target":"2025-Q4","window_end":"2025-12-31","shipped":null,"precision":null,"status":"late","slip_months_at_most":3,"quote":"Ironwood, our seventh generation TPU, will be generally available in the coming weeks.","source_url":"https://cloud.google.com/blog/products/compute/ironwood-tpus-and-new-axion-based-vms-for-your-ai-workloads","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v7/"},{"kind":"shipment","vendor":"Google","chip":"TPU7x (Ironwood)","field":"TPU7x generally available (GA)","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":null,"target":null,"window_end":null,"shipped":"2026-03-31","precision":"day","status":"late","slip_months_at_most":3,"quote":"March 31, 2026 Feature Generally available: TPU7x is generally available (GA).","source_url":"https://cloud.google.com/tpu/docs/release-notes","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/google-tpu-v7/"},{"kind":"figure","vendor":"AWS","chip":"Trainium2","field":"bf16","value":667,"unit":"TFLOPS","printed":"667","printed_unit":"TFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"delivered","slip_months_at_most":null,"quote":"667 BF16/FP16/TF32 TFLOPS","source_url":"https://awsdocs-neuron.readthedocs-hosted.com/en/latest/about-neuron/arch/neuron-hardware/trainium2.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/aws-trainium2/"},{"kind":"figure","vendor":"AWS","chip":"Trainium2","field":"fp8","value":1299,"unit":"TFLOPS","printed":"1,299","printed_unit":"TFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"delivered","slip_months_at_most":null,"quote":"1,299 FP8 TFLOPS","source_url":"https://awsdocs-neuron.readthedocs-hosted.com/en/latest/about-neuron/arch/neuron-hardware/trainium2.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/aws-trainium2/"},{"kind":"figure","vendor":"AWS","chip":"Trainium2","field":"memory","value":96,"unit":"GiB","printed":"96","printed_unit":"GiB","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"delivered","slip_months_at_most":null,"quote":"96 GiB of device memory with 2.9 TB/sec of bandwidth","source_url":"https://awsdocs-neuron.readthedocs-hosted.com/en/latest/about-neuron/arch/neuron-hardware/trainium2.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/aws-trainium2/"},{"kind":"figure","vendor":"AWS","chip":"Trainium2","field":"bandwidth","value":2.9,"unit":"TB/s","printed":"2.9","printed_unit":"TB/s","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"delivered","slip_months_at_most":null,"quote":"96 GiB of device memory with 2.9 TB/sec of bandwidth","source_url":"https://awsdocs-neuron.readthedocs-hosted.com/en/latest/about-neuron/arch/neuron-hardware/trainium2.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/aws-trainium2/"},{"kind":"shipment","vendor":"AWS","chip":"Trainium2","field":"Trn2 instances generally available","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":null,"target":null,"window_end":null,"shipped":"2024-12-03","precision":"day","status":"delivered","slip_months_at_most":null,"quote":"Today, AWS announces the general availability of Amazon Elastic Compute Cloud (Amazon EC2) Trn2 instances","source_url":"https://aws.amazon.com/about-aws/whats-new/2024/12/amazon-ec2-trn2-instances-available","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/aws-trainium2/"},{"kind":"figure","vendor":"AWS","chip":"Trainium3","field":"bf16","value":671,"unit":"TFLOPS","printed":"671","printed_unit":"TFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"671 BF16/FP16/TF32 TFLOPS","source_url":"https://awsdocs-neuron.readthedocs-hosted.com/en/latest/about-neuron/arch/neuron-hardware/trainium3.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/aws-trainium3/"},{"kind":"figure","vendor":"AWS","chip":"Trainium3","field":"fp8","value":2517,"unit":"TFLOPS","printed":"2,517","printed_unit":"TFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"2,517 MXFP8/MXFP4 TFLOPS","source_url":"https://awsdocs-neuron.readthedocs-hosted.com/en/latest/about-neuron/arch/neuron-hardware/trainium3.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/aws-trainium3/"},{"kind":"figure","vendor":"AWS","chip":"Trainium3","field":"fp4","value":2517,"unit":"TFLOPS","printed":"2,517","printed_unit":"TFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"2,517 MXFP8/MXFP4 TFLOPS","source_url":"https://awsdocs-neuron.readthedocs-hosted.com/en/latest/about-neuron/arch/neuron-hardware/trainium3.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/aws-trainium3/"},{"kind":"figure","vendor":"AWS","chip":"Trainium3","field":"memory","value":144,"unit":"GiB","printed":"144","printed_unit":"GiB","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"144 GiB of device memory, with 4.9 TB/sec of bandwidth","source_url":"https://awsdocs-neuron.readthedocs-hosted.com/en/latest/about-neuron/arch/neuron-hardware/trainium3.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/aws-trainium3/"},{"kind":"figure","vendor":"AWS","chip":"Trainium3","field":"bandwidth","value":4.9,"unit":"TB/s","printed":"4.9","printed_unit":"TB/s","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"144 GiB of device memory, with 4.9 TB/sec of bandwidth","source_url":"https://awsdocs-neuron.readthedocs-hosted.com/en/latest/about-neuron/arch/neuron-hardware/trainium3.html","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/aws-trainium3/"},{"kind":"promise","vendor":"AWS","chip":"Trainium3","field":"first Trainium3 instances available in late 2025","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2024-12-03","target":"2025-H2","window_end":"2025-12-31","shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"The first Trainium3-based instances are expected to be available in late 2025.","source_url":"https://press.aboutamazon.com/2024/12/aws-trainium2-instances-now-generally-available","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/aws-trainium3/"},{"kind":"shipment","vendor":"AWS","chip":"Trainium3","field":"Trn3 UltraServers generally available","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":null,"target":null,"window_end":null,"shipped":"2025-12-02","precision":"day","status":"kept","slip_months_at_most":0,"quote":"AWS announces the general availability of Amazon Elastic Compute Cloud (Amazon EC2) Trn3 UltraServers powered by our fourth-generation AI chip Trainium3","source_url":"https://aws.amazon.com/about-aws/whats-new/2025/12/amazon-ec2-trn3-ultraservers/","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/aws-trainium3/"},{"kind":"figure","vendor":"Intel","chip":"Gaudi 2","field":"bf16","value":432,"unit":"TFLOPS","printed":"432","printed_unit":"TFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"BF16 MME TFLOPS 432 1678","source_url":"https://cdrdv2-public.intel.com/817486/gaudi-3-ai-accelerator-white-paper.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/intel-gaudi2/"},{"kind":"figure","vendor":"Intel","chip":"Gaudi 2","field":"fp8","value":865,"unit":"TFLOPS","printed":"865","printed_unit":"TFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"FP8 MME TFLOPS 865 1678","source_url":"https://cdrdv2-public.intel.com/817486/gaudi-3-ai-accelerator-white-paper.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/intel-gaudi2/"},{"kind":"figure","vendor":"Intel","chip":"Gaudi 2","field":"memory","value":96,"unit":"GB","printed":"96","printed_unit":"GB","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"MEMORY 96GBHBM2E TDP 600W","source_url":"https://habana.ai/wp-content/uploads/2023/10/HL-225H_Datasheet_10_23.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/intel-gaudi2/"},{"kind":"figure","vendor":"Intel","chip":"Gaudi 2","field":"bandwidth","value":2.4,"unit":"TB/s","printed":"2.4","printed_unit":"TB/s","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"supporting extremely high memory capacity of 96GB and total throughput of 2.4TB/s","source_url":"https://habana.ai/wp-content/uploads/2023/10/HL-225H_Datasheet_10_23.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/intel-gaudi2/"},{"kind":"figure","vendor":"Intel","chip":"Gaudi 2","field":"power","value":600,"unit":"W","printed":"600","printed_unit":"W","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"supports card level TDP of 600 watts","source_url":"https://habana.ai/wp-content/uploads/2023/10/HL-225H_Datasheet_10_23.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/intel-gaudi2/"},{"kind":"promise","vendor":"Intel","chip":"Gaudi 2","field":"Supermicro Gaudi2 training server to market","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2022-05-10","target":"2022-H2","window_end":"2022-12-31","shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"Habana is partnering with Supermicro to bring the Supermicro Gaudi®2 Training Server to market in 2H 2022","source_url":"https://habana.ai/press/habana-labs-launches-second-generation-ai-deep-learning-processors/","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/intel-gaudi2/"},{"kind":"shipment","vendor":"Intel","chip":"Gaudi 2","field":"launched in Q2 2022 (Intel's word; no readable Intel page gives a first-shipment date)","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":null,"target":null,"window_end":null,"shipped":"2022-04","precision":"quarter","status":"kept","slip_months_at_most":0,"quote":"In Q2 2022, we launched our second-generation deep learning processors for training and inference: Habana Gaudi2* and Habana Greco*","source_url":"https://www.sec.gov/Archives/edgar/data/50863/000005086323000006/intc-20221231.htm","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/intel-gaudi2/"},{"kind":"figure","vendor":"Intel","chip":"Gaudi 3","field":"bf16","value":1678,"unit":"TFLOPS","printed":"1678","printed_unit":"TFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":3,"quote":"BF16 MME TFLOPS 432 1678","source_url":"https://cdrdv2-public.intel.com/817486/gaudi-3-ai-accelerator-white-paper.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/intel-gaudi3/"},{"kind":"figure","vendor":"Intel","chip":"Gaudi 3","field":"fp8","value":1678,"unit":"TFLOPS","printed":"1678","printed_unit":"TFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":3,"quote":"FP8 MME TFLOPS 865 1678","source_url":"https://cdrdv2-public.intel.com/817486/gaudi-3-ai-accelerator-white-paper.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/intel-gaudi3/"},{"kind":"figure","vendor":"Intel","chip":"Gaudi 3","field":"memory","value":128,"unit":"GB","printed":"128","printed_unit":"GB","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":3,"quote":"128 GB of HBM2e memory capacity, and 3.7 TB/s of HBM bandwidth","source_url":"https://cdrdv2-public.intel.com/817486/gaudi-3-ai-accelerator-white-paper.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/intel-gaudi3/"},{"kind":"figure","vendor":"Intel","chip":"Gaudi 3","field":"bandwidth","value":3.7,"unit":"TB/s","printed":"3.7","printed_unit":"TB/s","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":3,"quote":"128 GB of HBM2e memory capacity, and 3.7 TB/s of HBM bandwidth","source_url":"https://cdrdv2-public.intel.com/817486/gaudi-3-ai-accelerator-white-paper.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/intel-gaudi3/"},{"kind":"figure","vendor":"Intel","chip":"Gaudi 3","field":"power","value":900,"unit":"W","printed":"900","printed_unit":"W","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"late","slip_months_at_most":3,"quote":"supports up to 900W Total Device Power (TDP) with passive cooling","source_url":"https://cdrdv2-public.intel.com/817486/gaudi-3-ai-accelerator-white-paper.pdf","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/intel-gaudi3/"},{"kind":"promise","vendor":"Intel","chip":"Gaudi 3","field":"availability to server makers","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2024-04-09","target":"2024-Q2","window_end":"2024-06-30","shipped":null,"precision":null,"status":"late","slip_months_at_most":3,"quote":"Intel Gaudi 3 will be available to OEMs - including Dell Technologies, Hewlett Packard Enterprise, Lenovo and Supermicro - in the second quarter of 2024","source_url":"https://www.intc.com/news-events/press-releases/detail/1689/intel-unleashes-enterprise-ai-with-gaudi-3-ai-open-systems","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/intel-gaudi3/"},{"kind":"promise","vendor":"Intel","chip":"Gaudi 3","field":"launch","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2024-08-01","target":"2024-Q3","window_end":"2024-09-30","shipped":null,"precision":null,"status":"late","slip_months_at_most":3,"quote":"The Intel® Gaudi® 3 AI accelerator is also on track to launch in the third quarter","source_url":"https://www.sec.gov/Archives/edgar/data/50863/000005086324000122/q224_earningsrelease.htm","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/intel-gaudi3/"},{"kind":"shipment","vendor":"Intel","chip":"Gaudi 3","field":"launched in Q3 2024 (Intel's word)","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":null,"target":null,"window_end":null,"shipped":"2024-07","precision":"quarter","status":"late","slip_months_at_most":3,"quote":"Intel also launched its Intel® Gaudi® 3 AI accelerators","source_url":"https://www.sec.gov/Archives/edgar/data/50863/000005086324000147/q324_earningsrelease.htm","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/intel-gaudi3/"},{"kind":"shipment","vendor":"Huawei","chip":"Ascend 910C","field":"launched in the Atlas 900 A3 SuperPoD","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":null,"target":null,"window_end":null,"shipped":"2025-03","precision":"month","status":"delivered","slip_months_at_most":null,"quote":"In March 2025, Huawei officially launched the Atlas 900 A3 SuperPoD, which packs up to 384 Ascend 910C chips.","source_url":"https://www.huawei.com/en/news/2025/9/hc-xu-keynote-speech","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/huawei-ascend-910c/"},{"kind":"figure","vendor":"Huawei","chip":"Ascend 950","field":"fp8","value":1000,"unit":"TFLOPS","printed":"1","printed_unit":"PFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"They will be able to deliver 1 PFLOPS in FP8, MXFP8, and HiF8, and 2 PFLOPS in MXFP4.","source_url":"https://www.huawei.com/en/news/2025/9/hc-xu-keynote-speech","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/huawei-ascend-950/"},{"kind":"figure","vendor":"Huawei","chip":"Ascend 950","field":"fp4","value":2000,"unit":"TFLOPS","printed":"2","printed_unit":"PFLOPS","basis":"dense","per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"They will be able to deliver 1 PFLOPS in FP8, MXFP8, and HiF8, and 2 PFLOPS in MXFP4.","source_url":"https://www.huawei.com/en/news/2025/9/hc-xu-keynote-speech","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/huawei-ascend-950/"},{"kind":"promise","vendor":"Huawei","chip":"Ascend 950","field":"Ascend 950DT available","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2025-09-18","target":"2026-Q4","window_end":"2026-12-31","shipped":null,"precision":null,"status":"kept","slip_months_at_most":0,"quote":"The Ascend 950DT chip will be available in the fourth quarter of 2026.","source_url":"https://www.huawei.com/en/news/2025/9/hc-xu-keynote-speech","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/huawei-ascend-950/"},{"kind":"shipment","vendor":"Huawei","chip":"Ascend 950","field":"the Atlas 950 SuperPoD, built with Ascend 950DT chips, in large-scale commercial use","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":null,"target":null,"window_end":null,"shipped":"2026-09-17","precision":"day","status":"kept","slip_months_at_most":0,"quote":"Atlas 950 SuperPoD is seeing large-scale commercial use.","source_url":"https://www.huawei.com/en/news/2026/9/hc-wang-keynote","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/huawei-ascend-950/"},{"kind":"figure","vendor":"Microsoft","chip":"Maia 100","field":"memory","value":64,"unit":"GB","printed":"64","printed_unit":"GB","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"unrecorded","slip_months_at_most":null,"quote":"provide a total of 1.8 terabytes per second of bandwidth and 64 gigabytes of capacity","source_url":"https://techcommunity.microsoft.com/blog/azureinfrastructureblog/inside-maia-100-revolutionizing-ai-workloads-with-microsofts-custom-ai-accelerator/4229118","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/microsoft-maia-100/"},{"kind":"figure","vendor":"Microsoft","chip":"Maia 100","field":"bandwidth","value":1.8,"unit":"TB/s","printed":"1.8","printed_unit":"TB/s","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"unrecorded","slip_months_at_most":null,"quote":"provide a total of 1.8 terabytes per second of bandwidth and 64 gigabytes of capacity","source_url":"https://techcommunity.microsoft.com/blog/azureinfrastructureblog/inside-maia-100-revolutionizing-ai-workloads-with-microsofts-custom-ai-accelerator/4229118","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/microsoft-maia-100/"},{"kind":"figure","vendor":"Microsoft","chip":"Maia 100","field":"power","value":700,"unit":"W","printed":"700","printed_unit":"W","basis":null,"per_chips":1,"stated":null,"target":null,"window_end":null,"shipped":null,"precision":null,"status":"unrecorded","slip_months_at_most":null,"quote":"Designed to support up to 700W TDP but provisioned at 500W","source_url":"https://techcommunity.microsoft.com/blog/azureinfrastructureblog/inside-maia-100-revolutionizing-ai-workloads-with-microsofts-custom-ai-accelerator/4229118","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/microsoft-maia-100/"},{"kind":"promise","vendor":"Microsoft","chip":"Maia 100","field":"Maia and Cobalt chips start to roll out to Microsoft datacenters early in 2024","value":null,"unit":null,"printed":null,"printed_unit":null,"basis":null,"per_chips":null,"stated":"2023-11-15","target":"2024-H1","window_end":"2024-06-30","shipped":null,"precision":null,"status":"unrecorded","slip_months_at_most":null,"quote":"The chips will start to roll out early next year to Microsoft's datacenters","source_url":"https://news.microsoft.com/source/features/ai/in-house-chips-silicon-to-service-to-meet-ai-demand/","license":"quoted for comment","page":"https://jesseahawkins.com/chips/accelerators/microsoft-maia-100/"}]}