For 3 years, Tachyum has touted its Prodigy common processor with guarantees that it might be higher than CPUs and GPUs from the likes of AMD, Intel, and Nvidia. Nonetheless, the chip continues to be not in manufacturing regardless that it was initially projected to be in full quantity manufacturing again in 2021. Now it seems like the corporate is wanting forward and touting its Prodigy 2-based supercomputer design that guarantees 20 ExaFLOPS efficiency in 2025 ~ 2026.
Tachyum’s supercomputer design is supposed to ship 20 FP64 vector ExaFLOPS and 10 AI (INT8 or FP8) ZetaFLOPS efficiency inside an influence goal of 60MW in a footprint of 6,000 sq. toes in 2025, the corporate mentioned. The machine will use 64 Prodigy 2-based cupboards and 16 storage racks, however Tachyum did what number of Prodigy 2 processors it might have to ship this a lot efficiency.
Tachyum additionally mentioned it might construct a supercomputer providing 24.9 FP64 vector ExaFLOPS and 13.27 AI ZettaFLOPS consuming 73.8MW. To place the quantity into context, the upcoming El Capitan supercomputer powered by AMD’s Intuition MI300 datacenter APUs is about to ship round 2 FP64 ExaFLOPS.
Final 12 months the U.S. Division of Power expressed curiosity in procuring a 20 ExaFLOPS supercomputer with energy consumption of 20MW-60MW by 2025. Tachyum mentioned it had submitted a proposal to assemble such a system by 2025 however didn’t elaborate on the small print. Because the firm’s unique Prodigy didn’t meet Tachyum’s efficiency objectives (which is why it sued Cadence), it was affordable to imagine the corporate would use its 2nd-Gen Prodigy (which the corporate had talked about earlier than) for the machine.
This supercomputer design revealed by Tachyum this week primarily complies with the DoE’s necessities although the corporate doesn’t disclose the efficiency ranges it expects from its Prodigy 2.
Tachyum says its Prodigy processor is the first-ever common processor that may deal with a variety of demanding computing workloads. The unique Prodigy processor accommodates 128 proprietary 64-bit VLIW cores with two 1024-bit vector items and one 4096-bit matrix unit per core. The flagship Prodigy T16128-AIX processor was anticipated to offer a most of 90 FP64 teraflops for high-performance computing (HPC) and as much as 12 ‘AI petaflops’ for AI inference and coaching (with INT8 or FP8 precision). As well as, every chip was projected to have a energy consumption of as much as 950W and make use of liquid cooling.
Tachyum’s Prodigy 2 is anticipated to extend the variety of cores, and add HBM3, PCIe 6.0 + CXL, and built-in networking capabilities.
“Tachyum gives modern options from silicon to finish methods to handle the ever-increasing calls for for each HPC and AI,” mentioned Radoslav Danilak, founder and CEO of Tachyum. “Tachyum-designed supercomputers push the forefront of HPC efficiency whereas crossing the zetta-scale barrier for AI, reworking knowledge facilities into common computing facilities.”