Tachyum’s First CPUs Are Late, But It Plans 20 ExaFLOPS Supercomputer With Next-Gen CPUs Anyway
For three years, Tachyum has been promoting its Prodigy universal processor, promising to outperform CPUs and GPUs from the likes of AMD, Intel, and Nvidia. However, even though the chip was manufactured, it still isn’t. Originally, full-scale production was scheduled to begin in 2021Now the company looks forward. Selling a Prodigy 2-based supercomputer design It promises a performance of 20 ExaFLOPS in 2025-2026.
of Tachium supercomputer design According to the company, it aims to deliver a performance of 20 FP64 vector ExaFLOPS and 10 AI (INT8 or FP8) ZetaFLOPS within a power target of 60MW in a 6,000 square foot footprint in 2025. The machine uses 64 of his Prodigy 2-based cabinets and 16 of his racks of storage, but Tachyum calculated how many of his Prodigy 2 processors would be needed to achieve this much performance. .
Tachyum also says it can build a supercomputer that delivers 73.8MW consuming 24.9 FP64 vector ExaFLOPS and 13.27 AI ZettaFLOPS. To put the numbers into context, the upcoming El Capitan supercomputer powered by AMD’s Instinct MI300 data center APU is set to deliver around 2 FP64 ExaFLOPS.
Last year, the U.S. Department of Energy expressed interest in procuring 20 ExaFLOPS supercomputers with power consumption ranging from 20MW to 60MW by 2025. The company’s original his Prodigy failed to meet his Tachyum’s performance goals (which is why it sued Cadence), so the company decided to put a second-generation version of his Prodigy (which the company had previously talked about) into the machine. It was reasonable to assume it would.
The supercomputer’s design, revealed this week by Tachyum, is inherently compliant with DoE requirements, but the company has not disclosed the level of performance it expects from Prodigy 2.
According to Tachyum, the company’s Prodigy processor is the first universal processor to handle a wide variety of demanding computing workloads. The original his Prodigy processor contains his 128 proprietary 64-bit VLIW cores with two 1024-bit vector units and one 4096-bit matrix unit per core. The flagship Prodigy T16128-AIX processor is expected to deliver up to 90 FP64 teraflops for high-performance computing (HPC) and up to 12 “AI petaflops” (INT8 or FP8 precision) for AI inference and training. was Additionally, each chip was expected to consume up to 950W of power and employ liquid cooling.
Tachyum’s Prodigy 2 is expected to increase the number of cores, add HBM3, PCIe 6.0 + CXL, and integrated networking capabilities.
Radoslav Danilak, Founder and CEO of Tachyum, said: “Tachyum-designed supercomputers push the forefront of HPC performance while breaking the zettascale barriers of AI to transform data centers into universal computing centers.”