This week, Intel and the Barcelona Supercomputing Centre (BSC) mentioned they might make investments €400 million (round $426 million) in a laboratory that can develop RISC-V-based processors that could possibly be used to construct zettascale supercomputers. Nevertheless, the lab won’t focus solely on CPUs for next-generation supercomputers but in addition on processor makes use of for synthetic intelligence purposes and autonomous autos.
The analysis laboratory will presumably be arrange in Barcelona, Spain, and can obtain €400 million from Intel and the Spanish Authorities over 10 years. The basic function of the joint analysis laboratory is to develop chips primarily based on the open-source RISC-V instruction set structure (ISA) that could possibly be used for a variety of purposes, together with AI accelerators, autonomous autos, and high-performance computing.
The creation of the joint laboratory doesn’t mechanically imply that Intel will use RISC-V-based CPUs developed within the lab for its first-generation zettascale supercomputing platform however slightly signifies that the corporate is keen to make further investments in RISC-V. In any case, final yr, Intel tried to purchase SiFive, a number one developer of RISC-V CPUs and is among the many prime sponsors of RISC-V Worldwide, a non-profit group supporting the ISA.
Whereas round $21.3 million is a major sum of cash, Intel might be pouring much more into its x86-based merchandise within the coming years, so spending on RISC-V processors doesn’t imply a decrease concentrate on x86 designs. Quite the opposite, all through its historical past, Intel invested tons of of thousands and thousands in non-x86 architectures (together with RISC-based i960/i860 designs within the Eighties, Arm within the 2000s, and VLIW-based IA64/Itanium within the Nineties and the 2000s). Finally, these architectures have been dropped, however applied sciences developed for them discovered their method into x86 choices.
With its RISC-V efforts, Intel could possibly be killing a number of birds with one stone. First, suppose engineers from the joint laboratory handle to design a CPU know-how that’s extra appropriate for ZettaFLOPS-class supercomputers. In that case, Intel will be capable to use it for its merchandise. As an added bonus, Intel’s Foundry Companies division will seemingly grow to be a fab of alternative for CPUs/SoCs developed within the joint lab.
“Excessive-performance computing is the important thing to fixing the world’s most difficult issues, and we at Intel have an bold aim to dash to zettascale period for HPC,” mentioned Jeff McVeigh, vp and basic supervisor of the Tremendous Compute Group at Intel. “Barcelona Supercomputing Heart shares our imaginative and prescient for this aim, with equal emphasis on sustainability and an open method. We’re excited to companion with them to embark on this journey.”
Final yr Intel set itself an bold aim to construct a ZettaFLOPS-class supercomputer platform by 2027, which implies to extend the efficiency of supercomputers by 1000 occasions in about 5 years. The corporate mentioned it might want new compute architectures, new system architectures, high-speed reminiscence and I/O interfaces, novel fabrication applied sciences, and complicated chip packaging strategies, amongst different issues. One of many firm’s methods to radically enhance compute efficiency is to construct an structure that may mix the corporate’s x86 general-purpose cores with Xe-HPC compute GPUs. The primary product that makes use of this idea is Falcon Shores (opens in new tab) — which is already in improvement.