The TSMC 5nm process used by Intel's Gaudi 3 is expected to compete with other companies. Next year, Intel will use TSMC's 5nm Node for Gaudi 3 Accelerator.
Pat Gelsinger gave a small glimpse at the company's next-gen artificial intelligence accelerator, without giving any details about it. The Korean media has reported that Team Blue's Gaudi 3 is expected to compete with both the H 100 and the Instinct MI300X in terms of performance.
The Gaudi 3 will offer 1.5 times higher performance than the previous one, along with a four times increment in BFloat16 performance, twice as much compute, and a 50% increase in memory capacities, which will give us 144GB versus 96GB on the current. 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. The Gaudi3 is expected to launch next year and will be a single product.
The new information shows that Intel's Gaudi 3 will be based on TSMC's process. The H 100 and the MI300X are both based on a similar process, albeit with a slightly different variant known as 4N. The upcoming Goya inference chip will be designed with the help of the Taiwan-based Al chip Technologies. However, we are yet to see much about Intel's Gaudi architecture going forward, but by the looks of it, the future looks great.
Intel has stated that it is confident in the approach it will take towards the artificial intelligence markets going forward. It will be interesting to see how the Gaudi platform shapes up as Team Blue prepares to launch the next-gen accelerator.