Intel Gaudi 3 AI Accelerator Reportedly Built On TSMC 5nm Node, Faster Than NVIDIA H100
Intel's Gaudi 3 AI accelerator reportedly employs the TSMC 5nm process and is expected to compete with NVIDIA H100 & AMD MI300X.
Alongside the unveiling of Intel's 5th-Gen Xeon Scalable and Core Ultra "Meteor Lake" CPUs, CEO Pat Gelsinger provided a small glimpse at the company's next-gen AI accelerator, without giving out specifics about it. However, reports from the Korean media have revealed what Team Blue's Gaudi 3 accelerator could bring on board, and it is expected to compete with NVIDIA's H100 and AMD's Instinct MI300X AI GPUs in terms of AI performance.
Intel did reveal that its Gaudi 3 AI accelerator will offer 1.5 times higher performance than its previous counterpart, the Gaudi 2, along with a four times increment in BFloat16 performance, twice as much compute, and a 50% increase in memory capacities which gives us 144 GB versus 96GB on the current Gaudi2 accelerators. Moreover, it is expected that the accelerator will feature either HBM3 or HBM3e, which is expected to put it on par with current industry offerings, making it much more competitive and a viable alternative, especially when considering its perf/watt against NVIDIA's H100 GPU.
Our Gaudi roadmap remains on track with Gaudi3 out of the fab, now in packaging and expected to launch next year. In 2025, Falcon Shores brings our GPU and Gaudi capabilities into a single product.
Pat Gelsinger (Intel CEO) via SeekingAlpha
The newly surfaced information reveals that Intel's Gaudi 3 will be based on TSMC's 5nm process. NVIDIA's H100 makes use of a similar process, albeit a slightly more optimized variant known as 4N while the MI300X GPU employs both 5nm and 6nm process technologies.
Moreover, it is said that Intel has taken aid from the Taiwan-based Al chip
Read more on wccftech.com
