Tensordyne claims 13x throughput improvement over Nvidia's NVL72 GB300 with new TDN AI Processor

Tensordyne makes a big bet on log math to beat Nvidia

Tensordyne taped out Napier using logarithmic math, claiming 17x tokens per watt over Blackwell. For ops managing LLM inference at scale, it's a credible Nvidia alternative: 60% lower power, air-cooled, same density—cutting data center OPEX.