Intel revealed that it will be powering one of the world's fastest AI Supercomputers using its 5th Gen Xeon CPUs & Gaudi 2 accelerators.
Intel Gaudi 2 Accelerators Land In World's 15th Fastest AI Supercomputer, Gaudi 3 & Falcon Shores Next In Line To Tackle NVIDIA
The Gaudi-powered supercomputer is intended for AI workloads, particularly for the firm Stability AI. While Intel hasn't revealed specific figures on the computing power, it did disclose that the supercomputer will feature a massive 4,000 Gaudi 2 AI accelerators, and will have Intel's upcoming 5th Gen Xeon Scalable processors. Not only is this the first implementation for the new Xeon CPUs, but it also provides an opportunity for Gaudi accelerators to show their potential to the market.
Speaking of potential, in a recent coverage, we revealed that Intel has managed to close the gap between Gaudi AI accelerators and its competitors like the NVIDIA H100 where the company does offer better TCO even if they can't reach the same performance potential of the Hopper GPUs.
This could prove to be a huge moment for Intel and its AI division since the market is currently deprived of AI accelerators/GPUs due to the immense demand coupled with "bottlenecks" within the supply chain. If Intel managed to deliver an adequate stream of accelerators at a competitive price, it could very well be challenging the likes of NVIDIA and AMD.
The utilization of Gaudi 2 accelerators by Stability AI could act as a catalyst for further integrations. It will be interesting to see how the 5th Gen Xeon Scalable processors perform in the AI domain, given that its predecessors (Sapphire Rapids) are known to be one of the best options for AI-heavy workloads.
With rapid genAI development, one should expect similar implementations in the future as well, given that innovation always brings diversity with it. Gaudi accelerators are paving to be a viable alternative for the AI industry, and it won't be long before we see its wide-scale adoption.
That's why Intel didn't stop at Gaudi 2 and is already working on Gaudi 3 which is next in line and will be utilizing a 5nm process node. The Gaudi 3 accelerators will offer up to 4x BFloat16 capabilities, 2x compute, 1.5x network bandwidth, and 1.5x HBM capacities (144 GB vs. 96 GB). The follow-up to Gaudi 3 will be 2025's Falcon Shores which combines Guadi AI and Intel's GPU prowess on the same package.
WccftechContinue reading/original-link]
