NVIDIA Launches Vera Rubin NVL72: 3.7x Performance Boost in MLPerf Inference
Summary:NVIDIA has unveiled the Vera Rubin NVL72, a new AI inference system that demonstrates exceptional performance in the MLPerf Inference v6.1 benchmarks. Compared to its predecessor, the GB300 NVL72, the Vera Rubin NVL72 delivers up to 3.7x higher throughput on the Qwen3-VL model and 2.5x higher on the DeepSeek-R1 model. This leap in performance is achieved through hardware-software codesign, including enhanced Tensor Cores, the Transformer Engine, and NVFP4 precision technology. The launch of Vera
NVIDIA has launched its latest AI inference system, the Vera Rubin NVL72, which showcases impressive performance in the MLPerf Inference v6.1 benchmarks. Key technical highlights include:
Key Technical Highlights
- Performance Boost: The Vera Rubin NVL72 delivers up to 3.7x higher throughput on the Qwen3-VL model and 2.5x higher on the DeepSeek-R1 model compared to the GB300 NVL72.
- Hardware Optimization: Enhanced Tensor Cores and the Transformer Engine accelerate the prefill and decode stages of inference.
- Precision Technology: The NVFP4 precision technique reduces the memory footprint of model weights, attention, and KV cache, increasing throughput while maintaining output quality.
- Software Optimization: Continuous software optimizations have led to a 1.6x performance improvement in v6.1 over v6.0, with further optimizations ongoing.
- Scalability: The system achieves 99% scaling efficiency in offline scenarios, with throughput growing nearly linearly as hardware is added.
Industry Impact
The launch of the Vera Rubin NVL72 underscores NVIDIA's prowess in hardware-software codesign and provides higher efficiency and lower costs for AI inference tasks. For enterprises and research institutions requiring large-scale AI inference, this system can significantly enhance productivity and reduce operational costs.
Developer Recommendations
- Hardware Selection: For applications demanding high-performance AI inference, the Vera Rubin NVL72 is a strong contender.
- Software Optimization: Developers can leverage NVIDIA's software tools, such as NVIDIA Dynamo and TensorRT-LLM, to further optimize model performance.
- Stay Updated: Given NVIDIA's commitment to continuous software optimization, developers should regularly check for updates to benefit from future performance improvements.
— END —Source: NVIDIA AI News (2026-09-16)
Tags: #NVIDIA #AI Inference #MLPerf #Vera Rubin NVL72 #Hardware Acceleration
Editorial & Fact-Checking Note: This article is compiled from primary research, official release documentation, and source papers by the ZICQ Newsroom pipeline with automated entity verification and human editorial review. If you notice any technical inaccuracy, please submit a correction via our corrections policy or email our editorial desk directly.
Community Comments