Chips · accelerators · AMD
AMD Instinct MI250X · Instinct MI200
684 gigaflops of dense BF16 compute per watt, at 560 W
Shipped 8 Nov 2021; shipped, no dated promise on record. Epoch AI names it at no AI data center and 7 GPU clusters.
Embed
The frame credits AMD, Epoch AI, AI data centers and Epoch AI, Data on GPU clusters and links back to its page. Preview it.
Cite
Each figure in AMD’s own words
one chip, dense compute; memory is high-bandwidth memory- BF16, dense
- 383 TFLOPS
Peak bfloat16 383 TFLOPs
AMD Instinct MI250X product page - INT8, dense
- 383 TOPS
Peak INT8 Performance 383 TOPs
AMD Instinct MI250X product page - Memory
- 128 GB
Dedicated Memory Size 128 GB
AMD Instinct MI250X product page - Memory bandwidth
- 3.2 TB/s
Peak Memory Bandwidth 3.2 TB/s
AMD Instinct MI250X product page - Maximum power
- 560 W
Thermal Design Power (TDP) 500W | 560W
AMD Instinct MI250X product page - Compute per watt
- 684 GFLOPS/Wdense BF16 over maximum power; 684 GFLOPS/W at INT8
The MI250X has no FP8, so INT8 is given. AMD gives two TDP figures (500W and 560W); the record uses the higher one. No readable AMD page gives an availability window before the launch day, when AMD said the MI250X was already available from HPE.
The shipment
each date links to the page it is quoted from- 8 Nov 2021 Shipped: available from HPE in the HPE Cray EX supercomputer.
The AMD MI250X accelerator is currently available from HPE in the HPE Cray EX Supercomputer
New AMD Instinct MI200 Series Accelerators press release
Named at no AI data center Epoch AI tracks
H100-equivalents and power are the whole site’s today, with every chip it holds68,672 chips in 7 GPU clusters
Epoch AI’s list of AI supercomputers, which stops in July 2025| Cluster | Owner | Country | Chips | First operational |
|---|---|---|---|---|
| Oak Ridge NL Frontier | US Department of Energy | United States of America | 37,632 | 2022-05-30 |
| Eni HPC6 | Eni | Italy | 13,888 | 2024-11-18 |
| EuroHPC LUMI | EuroHPC JU | Finland | 11,912 | 2022-09-15 |
| Microsoft Explorer-WUS3 | Microsoft | United States of America | 1,920 | 2023-06-01 |
| GENCI Adastra | GENCI | France | 1,352 | 2022-06-01 |
| KT Internal MI250 Cluster | KT | South Korea | 1,200 | 2022-12-15 |
| Pawsey Supercomputing Centre Setonix | Pawsey Supercomputing Centre | Australia | 768 | 2022-07-01 |
Epoch AI’s names for this chip: AMD Radeon Instinct MI250X.
Other AMD chips
- Instinct MI300X 1,733 gigaflops of dense BF16 compute per watt, at 750 W
- Instinct MI300A 1,290 gigaflops of dense BF16 compute per watt, at 760 W
- Instinct MI325X 1,300 gigaflops of dense BF16 compute per watt, at 1,000 W
- Instinct MI355X 1,786 gigaflops of dense BF16 compute per watt, at 1,400 W
- AMD’s own pages, each linked beside the figure or date it gives. Each page was read and each quote checked on 23 Oct 2026.
- Where it runs: Epoch AI, AI data centers and Epoch AI, GPU clusters, CC BY 4.0.