Tachyum has released details about its 2 nanometer (nm) Prodigy Universal Processor, reporting specifications and performance claims aimed at hyperscale data centers, AI infrastructure, and high-performance computing (HPC) environments. Tachyum states that the 2nm Prodigy processor lines—Prodigy Ultimate and Prodigy Premium—are designed to deliver significant increases in AI rack performance, with the Ultimate model reportedly achieving up to 21.3 times higher performance than Nvidia’s Rubin Ultra NVL576 and the Premium model reaching up to 25.8 times the AI rack performance of Nvidia’s Vera Rubin 144.
Prodigy is described as the first chip to exceed 1,000 petaFLOPS (PFLOPs) on inference, compared to Nvidia Rubin’s 50 PFLOPs. Tachyum notes that the chip upgrade to 2nm technology enables notable reductions in power consumption, an important factor since multiple chiplets (each with 256 custom 64-bit cores) are integrated into a single package. The Ultimate model integrates 1,024 high-performance cores, 24 DDR5 17.6 GT/s memory controllers, and 128 PCIe 7.0 lanes. The Premium version supports 16 DRAM channels, with 512 to 128 cores, scalable up to 16-socket systems, while entry-level models offer 8 or 4 DRAM controllers and between 128 to 32 cores.
According to Tachyum, Prodigy supports a broad range of data center applications—spanning AI, supercomputing, digital currency, cloud, hyperscale, big data analytics, and databases—due to its scalability and performance segmentation. System support includes native operating systems, compilers, libraries, and out-of-the-box AI frameworks. Prodigy also offers compatibility with unmodified Intel and AMD x86 binaries, enabling mixed workloads alongside native applications.
Additional technical offerings include open-sourcing all supporting software and making Tachyum’s memory technology available for industry licensing, utilizing standard components that reportedly provide a tenfold memory bandwidth increase using dual in-line memory modules (DIMMs). Tachyum also reports that its Tachyum Processing Unit (TPU) core and Tachyum AI (TAI) data types are licensable, and the company is working to open its Instruction Set Architecture (ISA).
Tachyum claims Prodigy achieves up to five times integer performance, up to sixteen times higher AI performance, eight times DRAM bandwidth, four times chip-to-chip and I/O bandwidth, and twice the power efficiency relative to previous products. All chip configurations aim for reduced cost per core, and the product line is underpinned by a recent $220 million investment to proceed toward tape-out.
Dr. Radoslav Danilak, founder and CEO of Tachyum, said, “With tape-out funding now secured after a long wait, the world’s first Universal Processor can proceed to production, designed to overcome the inherent limitations of today’s data centers,” “The distinct markets addressed by Prodigy are the AI, server, and HPC markets, requiring fast and efficient chips. Tachyum’s Prodigy Premium and Ultimate will supercharge workloads with superior performance at a lower cost than any other solution on the market.”
Full specifications and documentation for the newest Prodigy processor are available at Tachyum’s website.
Source: Tachyum
