NVIDIA Vera Rubin debuts across MLPerf training and agentic endpoints
NVIDIA says Vera Rubin led its debut MLPerf Training submissions and delivered more than tenfold generational throughput-per-watt gains in a provisional Kimi K3 agentic endpoint result. The platform also showed up to twice the per-GPU training performance of GB300 and 97 per cent scaling efficiency from 256 to 1,024 GPUs.