Tech
NVIDIA Claims More AI Capacity From Every Megawatt With DSX
NVIDIA showcased Vera Rubin and DSX systems designed to improve AI performance, throughput and power efficiency.
What happened
At the AI Infra Summit in Santa Clara, NVIDIA presented Vera Rubin systems, DSX power-management software and related platform partnerships for AI factories. Lambda reported 24% higher cluster-wide token throughput and 23% better performance per watt using DSX MaxLPS, running 19 nodes within the power budget of 16. NVIDIA said MaxLPS can enable up to 40% more GPU capacity within the same megawatt budget in suitable Vera Rubin deployments. Its preview MLPerf Inference results showed Vera Rubin NVL72 delivering up to 3.7 times GB300 NVL72 throughput. Emerald AI also demonstrated automated load reduction for Silicon Valley Power while protecting priority workloads.
Why it matters
The results frame power efficiency as a direct driver of AI capacity and inference economics. Operators could generate more tokens, add compute within existing power limits and adjust lower-priority workloads when the grid requires reduced demand.
Source: NVIDIA Blog