NVIDIA Dominates MLPerf Inference v6 Benchmarks
NVIDIA submitted MLPerf Inference v6.0 results on April 1, 2026, reporting that its Blackwell Ultra platform delivered the highest throughput and lowest token cost across the suite. The company cites up to 2.77x token/sec speedups versus GB200 NVL72 and ninefold training wins over the nearest competitor, driven by hardware and software co-design. These results affect large-scale inference and TCO calculations.
Key Points
- 1Reports show NVIDIA's Blackwell Ultra achieves up to 2.77x token/sec improvements on MLPerf v6.0
- 2Highlights broad workload coverage, adding DeepSeek-R1, GPT-OSS-120B, Mixtral and more, stressing inference
- 3Signals infrastructure advantage for large-scale deployments, lowering token cost and improving TCO metrics
Scoring Rationale
High score due to official MLPerf v6.0 submissions showing industry-wide performance leadership and broad workload coverage; credibility benefits from submitted benchmark results and NVIDIA's transparency. I gave moderate novelty since this is performance improvement rather than a new architecture, and slightly reduced for vendor-authored coverage with limited technical detail.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems