Intel Gaudi 3 · Gaudi

1,864 gigaflops of dense BF16 compute per watt, at 900 W

Shipped the third quarter of 2024; shipped after the first date promised, up to 3 months. Epoch AI names it at no AI data center and no GPU cluster. Intel across the site.

Embed

Gaudi 3

<iframe src="https://jesseahawkins.com/chips/embed/accelerators/intel-gaudi3/" title="Intel Gaudi 3: 1,864 gigaflops of dense BF16 compute per watt, at 900 W" width="100%" height="520" style="border:0;max-width:100%" loading="lazy"></iframe>

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

Cite

This page

Jesse Hawkins, “Intel Gaudi 3: 1,864 gigaflops of dense BF16 compute per watt, at 900 W”, jesseahawkins.com, data retrieved 23 Oct 2026. https://jesseahawkins.com/chips/accelerators/intel-gaudi3/. 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{hawkins2026chipsacceleratorsintelgaudi3,
  author = {Hawkins, Jesse},
  title = {{Intel Gaudi 3: 1,864 gigaflops of dense BF16 compute per watt, at 900 W}},
  year = {2026},
  howpublished = {\url{https://jesseahawkins.com/chips/accelerators/intel-gaudi3/}},
  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 Intel’s own words

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

BF16, dense: 1,678 TFLOPS BF16 MME TFLOPS 432 1678 Intel Gaudi 3 AI Accelerator technical paper (July 2025, V1 Rev. 3) (read from a table with a column for each product, whose head reads “Feature/Product Intel® Gaudi® 2 AI Accelerator Intel® Gaudi® 3 AI Accelerator”)
FP8, dense: 1,678 TFLOPS FP8 MME TFLOPS 865 1678 Intel Gaudi 3 AI Accelerator technical paper (July 2025, V1 Rev. 3) (read from a table with a column for each product, whose head reads “Feature/Product Intel® Gaudi® 2 AI Accelerator Intel® Gaudi® 3 AI Accelerator”)
Memory: 128 GB 128 GB of HBM2e memory capacity, and 3.7 TB/s of HBM bandwidth Intel Gaudi 3 AI Accelerator technical paper (July 2025, V1 Rev. 3)
Memory bandwidth: 3.7 TB/s 128 GB of HBM2e memory capacity, and 3.7 TB/s of HBM bandwidth Intel Gaudi 3 AI Accelerator technical paper (July 2025, V1 Rev. 3)
Maximum power: 900 W supports up to 900W Total Device Power (TDP) with passive cooling Intel Gaudi 3 AI Accelerator technical paper (July 2025, V1 Rev. 3)
Compute per watt: 1,864 GFLOPS/Wdense BF16 over maximum power; 1,864 GFLOPS/W at FP8

Figures are for the HL-325L OAM card (the PCIe card is 600W). Intel's April 2024 promise of OEM availability in Q2 2024 is on intc.com and intel.com; intel.com answers 403 and intc.com is not on the record's Intel host list, so it is left out.

2 dated promises and the shipment

each date links to the page it is quoted from

9 Apr 2024 Promised availability to server makers in the second quarter of 2024. Intel Gaudi 3 will be available to OEMs - including Dell Technologies, Hewlett Packard Enterprise, Lenovo and Supermicro - in the second quarter of 2024 Intel unleashes enterprise AI with Gaudi 3, AI open systems strategy and new customer wins
1 Aug 2024 Promised launch in the third quarter of 2024. The Intel® Gaudi® 3 AI accelerator is also on track to launch in the third quarter Intel second quarter 2024 financial results, 8-K exhibit 99.1
the third quarter of 2024 Shipped: launched in Q3 2024 (Intel's word). Intel also launched its Intel® Gaudi® 3 AI accelerators Intel third quarter 2024 financial results, 8-K exhibit 99.1

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 Intel chips

Gaudi 2 720 gigaflops of dense BF16 compute per watt, at 600 W

Sources

accelerators.csv Every dataset → Status → Glossary →

Intel’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.