Tachyum announced details and specifications for its 2 nm Prodigy Universal Processor, positioned to support Artificial Intelligence models with parameter counts many orders of magnitude larger than existing solutions while reducing cost. The company described the design as a universal processor and said the product family targets significantly higher rack efficiency for inference workloads compared with current alternatives.
In published comparisons, Tachyum said Prodigy Ultimate delivers up to 21.3x higher Artificial Intelligence rack performance than the NVIDIA Rubin Ultra NVL576, while Prodigy Premium delivers up to 25.8x higher Artificial Intelligence rack performance than the Vera Rubin 144. The announcement contrasts the Prodigy family with existing Rubin systems and presents the new chips as a substantial multiplier on rack-level inference throughput.
Tachyum also stated that the 2 nm Prodigy is the first chip to exceed 1,000 PFLOPs on inference and that technical details for the design will be published within a week. For context provided in the announcement, NVIDIA Rubin was cited as delivering 50 PFLOPs. The company emphasized the combination of process node, claimed inference performance, and cost efficiency as the primary differentiators for the Prodigy Universal Processor lineup.
