At the end of 2022, Intel officially released a new GPU Max series product, codenamed Ponte Vecchio, which is positioned in HPC high-performance computing and AI artificial intelligence, and is head-on with NVIDIA and AMD Instinct series, which is also part of Intel's XPU strategy.
However, after its birth, there are few applications of Intel GPU Max, and the only well-known is the American exascale supercomputer Aorura, with the fourth-generation Xeon Max series, but in the latest supercomputing list, it still has not beaten the Frontier of the AMD platform, ranking second.
据最新曝料,Intel已经告知合作伙伴,GPU Max系列将逐步停产退市,Intel AI加速器的重点将放在Guadi 2/3系列,还有明年的第二代GPU Max Falcon Shores。
The Ponte Vecchio GPU Max was a project promoted by Raja Koduri back then, using 5 different manufacturing processes, 47 different modules, more than 100 billion transistors, equipped with up to 128 Xe-HPC high-performance computing cores, 128 ray tracing cores, 64MB L1 cache, 108MB L2 cache, 128GB HBM high-bandwidth memory, and 600W of power consumption at full blood.
Gaudi 3 is an independent AI accelerator, which was just released not long ago, upgrading TSMC's 5nm process, equipped with 8 matrix engines, 64 tensor cores, 96MB SRAM cache, 1024-bit 128GB HBM2E memory, and a full blood power consumption of 900W.
Intel claims that it has 2x the FP8 AI computing power, 4 times the BF16 AI computing power, 2x the network bandwidth, 1.5 times the memory bandwidth, and is 50% ahead of NVIDIA H100 LLM inference performance and 40% faster training time than the previous generation.
For the Chinese market, the Gaudi 3 will be available in a special version with a maximum power consumption limit of 450W.
Falcon Shores最初规划融合CPU、GPU，就像AMD Instinct MI300A，但后来改为纯GPU方案，融合设计则后续再说。