B · Normal
[Nscale tests NVIDIA DSX solution and increases inference throughput by 49%] On October 21, NVIDIA disclosed that Nscale used DSX MaxLPS in the Iceland AI factory to evaluate four NVIDIA GB300 NVL72 racks. Under a fixed power supply budget of 264 kilowatts, the inference throughput increased by 49% and the manageable GPU capacity increased by 37%. The above results are from a specific test environment and do not represent actual performance in all deployment scenarios.
Comments