MLPerf's New Inference Benchmarks Put NVIDIA on Top, But...

Why It Matters

The rankings shape data‑centre purchasing decisions and signal which vendors are best positioned to meet escalating AI inference demands, influencing both market share and future R&D investment.

Key Takeaways

• Blackwell Ultra tops 9 of 12 MLPerf tests
• AMD's MI300X excels in vision workloads
• Intel Gaudi3 leads in language models
• Power efficiency gap narrows for competitors
• Benchmark updates mirror emerging transformer workloads

Pulse Analysis

MLPerf’s benchmark suite has become the de‑facto yardstick for measuring AI inference performance in enterprise settings. Version 6.0, launched in April 2026, broadens its test matrix to cover a wider array of transformer architectures, video analytics, and recommendation systems. By standardizing hardware, software stacks, and data‑set selections, the benchmark provides a level playing field that lets operators compare raw throughput, latency, and power draw across disparate platforms.

In the latest results, NVIDIA’s Blackwell Ultra GPU claimed top spots in nine of the twelve categories, delivering up to 2.3× higher throughput than its predecessor on large language‑model inference. AMD’s MI300X shone in computer‑vision tasks, while Intel’s Gaudi 3 secured the lead in natural‑language processing workloads, narrowing the performance gap that NVIDIA has traditionally enjoyed. Notably, power‑efficiency scores improved across the board, with AMD and Intel showing up to 15 % lower watts‑per‑inference compared to prior generations, a trend that could sway cost‑sensitive cloud providers.

The implications extend beyond bragging rights. Data‑centre operators use MLPerf data to justify capital expenditures, and a clear performance lead often translates into larger market share for GPU vendors. As transformer models continue to dominate AI workloads, vendors are likely to accelerate hardware optimizations and software ecosystem support to capture this growing segment. Meanwhile, the tighter efficiency margins suggest a shift toward greener AI deployments, aligning with corporate sustainability goals and potentially reshaping procurement strategies in the next fiscal cycle.