AMD Instinct MI325X · Instinct MI300

1,300 gigaflops of dense BF16 compute per watt, at 1,000 W

Promised for the fourth quarter of 2024; past its date, no dated shipment on record. Epoch AI names it at no AI data center and no GPU cluster.

Embed

Instinct MI325X

<iframe src="https://jesseahawkins.com/chips/embed/accelerators/amd-mi325x/" title="AMD Instinct MI325X: 1,300 gigaflops of dense BF16 compute per watt, at 1,000 W" width="100%" height="520" style="border:0;max-width:100%" loading="lazy"></iframe>

The frame credits AMD and Epoch AI, AI data centers and links back to its page. Preview it.

Cite

This page

Jesse Hawkins, “AMD Instinct MI325X: 1,300 gigaflops of dense BF16 compute per watt, at 1,000 W”, jesseahawkins.com, data retrieved 23 Oct 2026. https://jesseahawkins.com/chips/accelerators/amd-mi325x/. Data: Each vendor’s own datasheets and releases, quoted; Epoch AI, “AI data centers” and “Data on GPU clusters”, CC BY 4.0.

BibTeX

@misc{hawkins2026chipsacceleratorsamdmi325x,
  author = {Hawkins, Jesse},
  title = {{AMD Instinct MI325X: 1,300 gigaflops of dense BF16 compute per watt, at 1,000 W}},
  year = {2026},
  howpublished = {\url{https://jesseahawkins.com/chips/accelerators/amd-mi325x/}},
  note = {Data: Each vendor’s own datasheets and releases, quoted; Epoch AI, ``AI data centers'' and ``Data on GPU clusters'', CC BY 4.0, retrieved 2026-10-23}
}

The data · Epoch AI

Epoch AI, 'AI data centers'. Published online at epoch.ai. Retrieved from 'https://epoch.ai/data/ai-data-centers' [online resource]. CC BY 4.0.

The clusters · Epoch AI

Epoch AI, 'Data on GPU clusters'. Published online at epoch.ai. Retrieved from 'https://epoch.ai/data/gpu-clusters' [online resource]. CC BY 4.0.

CSV Post on X

Each figure in AMD’s own words

one chip, dense compute; memory is high-bandwidth memory

BF16, dense: 1,300 TFLOPS Peak bfloat16 1.3 PFLOPs AMD Instinct MI325X product page (1.3 PFLOPS on the page)
FP8, dense: 2,610 TFLOPS Peak Eight-bit Precision (FP8) Performance (E5M2, E4M3) 2.61 PFLOPs AMD Instinct MI325X product page (2.61 PFLOPS on the page)
Memory: 256 GB Dedicated Memory Size 256 GB AMD Instinct MI325X product page
Memory bandwidth: 6 TB/s Peak Memory Bandwidth 6 TB/s AMD Instinct MI325X product page
Maximum power: 1,000 W Typical Board Power (TBP) 1000W AMD Instinct MI325X product page
Compute per watt: 1,300 GFLOPS/Wdense BF16 over maximum power; 2,610 GFLOPS/W at FP8

No readable AMD page or filing states that MI325X production shipments started; AMD's earnings releases for Q4 2024 to Q2 2025 do not name the MI325X, so delivered is left empty. The June 2024 announcement gave 288GB of memory; the shipped part has 256 GB.

3 dated promises

each date links to the page it is quoted from

2 Jun 2024 Promised availability in the fourth quarter of 2024. The expanded roadmap includes the AMD Instinct MI325X accelerators, with planned availability in Q4 2024 AMD Computex 2024 press release
30 Jul 2024 Promised availability in the fourth quarter of 2024. The roadmap includes the new AMD Instinct MI325X accelerator, planned to be available in Q4 2024 AMD second quarter 2024 financial results, 8-K exhibit 99.1
10 Oct 2024 Promised production shipments in the fourth quarter of 2024. AMD Instinct MI325X accelerators are currently on track for production shipments in Q4 2024 AMD Instinct MI325X launch 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 holds

In no GPU cluster of Epoch AI’s list

Epoch AI’s list of AI supercomputers, which stops in July 2025

Other AMD chips

Instinct MI250X 684 gigaflops of dense BF16 compute per watt, at 560 W
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 MI355X 1,786 gigaflops of dense BF16 compute per watt, at 1,400 W

Sources

accelerators.csv Every dataset → Status → Glossary →

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.