1 of 1
Story summary
- MLCommons released MLPerf Inference v5.1 benchmarks, featuring NVIDIA's Blackwell Ultra GB300, AMD's Instinct MI355X, and Intel's Arc Pro B60.
- NVIDIA's GB300 outperformed the GB200 by 45%, setting records in benchmarks like DeepSeek-R1 and Llama 3.1.
- AMD's MI355X showed a 2.7x performance boost but faced limitations due to smaller test configurations.
- NVIDIA's new disaggregated serving techniques improved throughput and efficiency in AI inference.
- The inference market is expected to grow significantly, with NVIDIA retaining a leading role.
