Nvidia sets new MLPerf records with 288 GPUs while AMD and Intel focus on different battles

Editor's take

Nvidia has achieved new high scores on the MLPerf inference benchmark, leveraging a substantial cluster of 288 H100 GPUs, while competitors AMD and Intel are emphasizing different aspects of their performance.

This development underscores Nvidia's continued dominance in raw inference throughput for large-scale deployments, particularly as the benchmark now includes more demanding multimodal and video workloads. The divergence in reporting from AMD and Intel suggests they are strategically focusing on specific market segments or performance characteristics where their architectures might offer advantages, rather than directly competing on Nvidia's chosen metrics.

Future MLPerf rounds will be critical to observe if AMD's Instinct MI300X or Intel's Gaudi 2 and upcoming Gaudi 3 can close the gap on these new multimodal tasks, or if Nvidia's ongoing hardware and software optimizations continue to widen the lead. The industry will also be watching how cloud providers and enterprises prioritize cost-effectiveness and power efficiency versus peak performance as these benchmarks evolve.