NVIDIA’s DSX MaxLPS Delivers 40% More Token Throughput per Megawatt as Lambda Cluster Reaches 5 Million Tokens per Second

Wccftech reports that NVIDIA showcased energy-efficiency optimizations for its Vera Rubin platforms at the AI Infra Summit. The company says its DSX MaxLPS technology delivers 40% more token throughput per megawatt.

The report also says that a Lambda cluster has reached 5 million tokens per second. It presents the result as part of NVIDIA’s broader effort to address the energy demands created by AI workloads.

NVIDIA’s full-stack AI factory platform includes Vera Rubin systems, Dynamo inference software, NeMo libraries, and networking technologies. These include NVLink for scale-up computing, Spectrum-X for Ethernet, and ConnectX SuperNICs for connecting large numbers of systems.

The reported efficiency gains are presented as features of Vera Rubin platforms. The available information does not provide additional testing conditions or independent verification for the figures.