Tachyum Prodigy 2nm processor: specs, AI and HPC claims, NVIDIA comparisons

Tachyum, a company that has spent several years promising to upend the general‑purpose processor market, has unveiled refreshed specifications for its forthcoming Prodigy chip, built on a 2 nm process. The company still hasn’t brought a real product to market, yet the latest claims are anything but modest.

According to the developers, the 2 nm Prodigy would be the first processor to push past 1,000 PFLOPS on AI workloads. The company says the new generation delivers a fivefold jump in integer performance and up to 16x faster AI task processing. Versus today’s offerings, Tachyum promises Prodigy will be three times quicker than the best x86 CPUs in AI jobs and six times more powerful than the fastest GPGPUs in HPC. If those figures hold, the bar for compute‑class silicon shifts dramatically.

The architecture aims big. A full 2 nm Prodigy integrates 256 cores, while server systems are assembled from multiple such dies. The lineup will span three tiers:

Prodigy Ultimate — up to 768–1024 cores and 24 DDR5 channels, with top models reaching 6.0 GHz.

Prodigy Premium — 256–512 cores and 16 DDR5 channels with support for 16‑die configurations.

Entry‑level Prodigy — up to 128 cores and 4 or 8 DDR5 channels.

The boldest claims target NVIDIA’s server platforms. By the company’s internal estimates, systems based on Prodigy Premium surpass NVIDIA Vera Rubin NVL144 in AI performance by nearly 26x, while Prodigy Ultimate configurations outpace Rubin NVL576 by about 22x. Those are sweeping margins that, if replicated in practice, could reorder the field.

Tachyum says a recent $220 million investment is moving the project toward final production and a tape‑out for test manufacturing. If the company delivers even a slice of what it outlines, high‑performance computing could be staring at one of the biggest breakthroughs in years—though until silicon ships, expectations will remain measured.