Intel has revealed a shift in strategy that affects its XPU and data-center item roadmap.XPU is an effort by
Intel to integrate numerous pieces of silicon into one plan. The plan was to integrate CPU, GPU, networking, FPGA, and AI accelerator and usage software to pick the best processor for the job at hand.That’s an ambitious project, and it appears like Intel is confessing that it can’t do it, at least for now.Jeff McVeigh,
corporate vice president and basic manager of the Super Compute Group at Intel, supplied an update to the data-center processor roadmap that includes taking a few steps back. Its proposed mix CPU and GPU, code-named Falcon Shores, will now be a GPU chip only.
“A lot has changed in the previous 12 months. Generative AI is changing everything. And from our viewpoint, from Intel’s perspective, we feel it is premature to be integrating the CPU and GPU for the next-generation item,” McVeigh stated throughout a press rundown at the ISC High Efficiency Conference in Hamburg, Germany.The former strategy called for the CPU and GPU to be on the very same development cycle, but the GPU could take longer to establish than the CPU, which would have suggested the CPU innovation would sit idle while the GPU was being developed. Intel chose that the dynamic nature of today’s market dictates a requirement for discrete services.”I’ll confess, I was incorrect. We were moving too quick down the XPU path
. We feel that this dynamic nature will be much better served by having that versatility at the platform level. And after that we’ll incorporate when the time is right,”McVeigh said.The result is a substantial change in Intel’s roadmap. Intel in March ditched a supercomputer GPU codenamed Rialto Bridge, which was to be the follower to the Max Series GPU, codenamed Ponte Vecchio, which is already on the market.The brand-new Falcon Shores chip, which is the successor to Ponte Vecchio, will now be a next-generation discrete GPU targeted at both high-performance computing and AI. It consists of the AI processors, standard Ethernet switching, HBM3 memory, and I/O at scale, and
it is now due in 2025. McVeigh said that Intel hasn’t ruled out combining a CPU and GPU, but it’s not the priority today.” We will at the correct time … when the window of weather condition is right, we’ll do that. We simply don’t feel like it’s best in this next generation.” Other Intel news McVeigh also talked up enhancements to Intel’s OneAPI toolkit, a household of compilers, libraries, and programming tools that can perform code on the Xeon, Falcon Shores GPU, and Gaudi AI processor. Compose once, and the API can select the best chip on which to execute. The most recent upgrade provides speed gains for HPC applications with OpenMP GPU offload, extended support for OpenMP and Fortran, and sped up AI and deep learning. On the supercomputer front, Intel has delivered more than 10,624 calculate nodes of Xeon Max Series chips with HBM for the Aurora supercomputer, which includes 21,248 CPU nodes, 63,744 GPUs, 10.9 PB of DDR memory, and 230PB of storage. Aurora is being developed at the Lawrence Livermore National Labs and will exceed 2 exaFLOPs of efficiency when complete. When functional, it’s expected
to dethrone Frontier as the fastest supercomputer on the planet. Intel likewise talked about servers from Supermicro that appear to be focused on handling Nvidia’s … Source
