InferenceMAX: Open Source Inference Benchmarking

InferenceMAX v1 is an open-source, automated inference benchmark that reruns a suite of tests nightly on hundreds of chips to track how software improvements change real performance. The project publishes a free live dashboard at inferencemax.ai, and the initial v1 release benchmarks GB200 NVL72, B200, MI355X, H200, MI325X, H100 and MI300X. The article states that AMD and Nvidia GPUs each perform better on different workloads and interactivity levels, and it says the project plans to add Google TPU and AWS Trainium backends over the next two months.