Intel revises its XPU strategy

Intel has introduced a shift in strategy that impacts its XPU and data-center product or service roadmap.

XPU is an exertion by Intel to combine numerous pieces of silicon into 1 offer. The prepare was to combine CPU, GPU, networking, FPGA, and AI accelerator and use program to pick the most effective processor for the task at hand.

That is an ambitious undertaking, and it appears like Intel is admitting that it simply cannot do it, at least for now.

Jeff McVeigh, corporate vice president and common manager of the Tremendous Compute Team at Intel, presented an update to the info-middle processor roadmap that requires taking a couple methods again. Its proposed mixture CPU and GPU, code-named Falcon Shores, will now be a GPU chip only.

“A good deal has adjusted in the earlier 12 months. Generative AI is reworking all the things. And from our standpoint, from Intel’s standpoint, we experience it is premature to be integrating the CPU and GPU for the future-generation solution,” McVeigh claimed during a press briefing at the ISC High General performance Conference in Hamburg, Germany.

The former strategy termed for the CPU and GPU to be on the very same enhancement cycle, but the GPU could take for a longer time to establish than the CPU, which would have intended the CPU technological innovation would sit idle while the GPU was currently being produced. Intel made the decision that the dynamic mother nature of present day current market dictates a will need for discrete solutions.

“I’ll admit it, I was erroneous. We had been going as well rapidly down the XPU path. We come to feel that this dynamic mother nature will be better served by getting that versatility at the platform level. And then we will combine when the time is correct,” McVeigh said.

The end result is a considerable improve in Intel’s roadmap.

Intel in March scrapped a supercomputer GPU codenamed Rialto Bridge, which was to be the successor to the Max Collection GPU, codenamed Ponte Vecchio, which is presently on the industry.

The new Falcon Shores chip, which is the successor to Ponte Vecchio, will now be a next-generation discrete GPU specific at the two substantial-effectiveness computing and AI. It contains the AI processors, normal Ethernet switching, HBM3 memory, and I/O at scale, and it is now thanks in 2025.

McVeigh claimed that Intel hasn’t ruled out combining a CPU and GPU, but it can be not the priority suitable now. “We will at the correct time … when the window of climate is right, we’ll do that. We just really do not sense like it’s appropriate in this future technology.”

Other Intel news

McVeigh also talked up improvements to Intel’s OneAPI toolkit, a family members of compilers, libraries, and programming resources that can execute code on the Xeon, Falcon Shores GPU, and Gaudi AI processor. Produce once, and the API can select the very best chip on which to execute. The most up-to-date update delivers speed gains for HPC purposes with OpenMP GPU offload, prolonged assist for OpenMP and Fortran, and accelerated AI and deep discovering.

On the supercomputer entrance, Intel has delivered extra than 10,624 compute nodes of Xeon Max Series chips with HBM for the Aurora supercomputer, which involves 21,248 CPU nodes, 63,744 GPUs, 10.9PB of DDR memory, and 230PB of storage. Aurora is becoming developed at the Lawrence Livermore Nationwide Labs and will exceed 2 exaFLOPs of overall performance when total. When operational, it is really anticipated to dethrone Frontier as the speediest supercomputer in the globe.

Intel also talked over servers from Supermicro that feel to be aimed at using on Nvidia’s DGX AI techniques. They feature 8 Ponte Vecchio Max Sequence GPUs, each with 128 GB of HBM memory for additional than 1 TB overall of HBM memory per system. Not shockingly, the servers are specific at AI deployments. The solution is predicted to be broadly available in Q3.