NVIDIA Vera Rubin NVL72 Debuts with Major Performance Gains in MLPerf Inference v6.1

In its first MLPerf Inference preview submission, the NVIDIA Vera Rubin NVL72 system delivered up to 3.7x better throughput than the GB300 NVL72. [2] NVIDIA submitted these preview results on two demanding benchmarks in the MLPerf Inference v6.1 suite: DeepSeek-R1 and Qwen3-VL. [2] Using vLLM with the NVIDIA Dynamo open source inference framework, the Vera Rubin NVL72 delivered up to 3.7x higher throughput than the GB300 NVL72 on Qwen3-VL across offline, server, and interactive scenarios. [2] On DeepSeek-R1, using the NVIDIA TensorRT-LLM library, throughput was up to 2.5x higher than the GB300 NVL72. [2]

A 288-GPU submission across four GB300 NVL72 racks achieved 99% scaling efficiency, demonstrating nearly linear throughput growth from a single-rack baseline. [2] Additionally, software optimizations in NVIDIA’s MLPerf Inference v6.1 submissions resulted in up to 1.6x higher performance compared to v6.0. [2] Meanwhile, AI agents are expanding from cloud data centers to vehicles, robots, and other edge devices. [1]

The substantial throughput gains of the Vera Rubin NVL72, particularly its 3.7x advantage over the GB300 NVL72 on the Qwen3-VL benchmark, underscore rapid hardware advancements in handling demanding models. The near-linear scaling efficiency of the GB300 NVL72 suggests strong multi-rack operational viability. What the sources don't address: specific pricing or availability dates for the Vera Rubin NVL72 system.