Fujitsu has revealed that the corporate is prepping the successor for its A64FX processor for high-performance computing. The corporate’s second-generation Arm-based server CPU is slated to supply significantly greater efficiency and power effectivity than its predecessor, in addition to will add options to deal with AI and information analytics functions. The CPU is codenamed Monaka and it’ll arrive typically in 2027 and can energy a next-generation supercomputer due in 2028.
Like the unique A64FX, Fujitsu’s Monaka will as soon as once more be an Arm ISA processor. However it’s going to additionally combine {hardware} to speed up synthetic intelligence (AI) and information analytics functions, in response to particulars launched by the corporate at its ActivateNow: Know-how Summit on the Laptop Historical past Museum in Mountain View, California, experiences The Register.
The promise to spice up efficiency in conventional HPC and rising AI workloads is logical. Though Fujitsu’s present A64FX already has assist for 512-bit Scalable Vector Extensions (SVE) and may function in FP64, FP32, FP16 and INT8 modes for a wide range of AI and conventional supercomputer functions, the quickly growing area of AI workloads has been adopting new information codecs past FP16 and INT8. In the meantime, Retaining the Arm structure will make sure that the Monaka processor will be capable of run code developed for the unique A64FX CPU in addition to different Arm-based system-on-chips for datacenters.
“The subsequent-generation DC CPU (Monaka) that we’re growing can have a wider vary of options and can show extra power environment friendly,” a Fujitsu spokesperson informed The Register. “The vary of potential functions is wider than that of the A64FX, which has particular traits (e.g., interconnects) particular to Fugaku.
One in all Fujitsu’s principal targets with Monaka is to supply ‘overwhelming power effectivity’ when put next with competing processors obtainable on the time, claims The Register citing the corporate’s officers. The agency is aiming to ship 70% greater total efficiency and 100% greater performance-per-watt than competing chips. Although with supply not anticipated till 2027, it goes with out saying that any aggressive efficiency expectations are aspirational at greatest.
Fujitsu’s present 48+4-core A64FX processor for HPC has confirmed that the Arm structure is completely able to powering supercomputers, on this case Fugaku, which was the world’s quickest supercomputer from 2020 to 2022. However the CPU is mainly tailor-made for conventional supercomputer workloads, and consequently it is solely been utilized in a handful of programs, together with Fugaku, Fujitsu’s PrimeHPC FX700 and FX1000 programs (which can be found for buy), and HPE’s Apollo 80 HPC platform.
Monaka, in flip, will enable Fujitsu to take a stab at supplying the broader HPC market with a excessive efficiency Arm processor. Whereas the corporate is not providing particular technical particulars presently, they’re making it clear that they are designing the chip for a wider viewers, versus the supercomputer-focused A64FX and its area of interest options like on-package HBM2 and the Tofu Interconnect D cloth to attach a number of nodes in a cluster. Shifting to a broader viewers opens up extra gross sales alternatives for Fujitsu, however it’s going to put the corporate in additional direct competitors with different Arm server CPU distributors comparable to NVIDIA, Ampere, and the numerous inside tasks at hyperscalers.
In any case, it will be attention-grabbing to see how issues unfold as soon as Monaka arrives in 2027. The Arm server CPU market has shortly blossomed over the previous few years, so by the point Monaka hits the scene, it may be coming right into a market with numerous alternative for Arm servers and Arm software program, but in addition a market with no scarcity of firms attempting to say their piece of the pie.
