XPerf says MiTAC AI rack validation ran 10x faster and beat AMD benchmarks

XPerf announced a deployment with MiTAC Computing that it says validated an AMD Instinct MI325X AI rack in days, not weeks, while delivering 20% better performance than AMD’s publicly reported FLUX.1-dev benchmark under the tested setup. The companies said the results point to faster, more revenue-linked rack validation for AI operators, with full MLPerf Training V6.0 results now submitted.

Why it matters: - XPerf is pitching ClusterReady™ as a faster way to prove AI rack performance before deployment. - The software targets the gap between basic rack testing and sustained production workloads that generate revenue. - XPerf says the approach can cut validation time from weeks to hours or days. - MiTAC Computing says the results highlight its ability to build high-performance AI infrastructure.

What happened: - XPerf announced a deployment with MiTAC Computing on June 18, 2026. - ClusterReady™ validated an AMD Instinct™ MI325X AI air-cooled rack using MiTAC Computing G8825Z5 servers. - The tested rack delivered performance 20% higher than AMD’s publicly reported FLUX.1-dev benchmark results under the tested configuration. - The test produced the first reported Llama 3.1 8B pre-training performance results on MI325X. - XPerf said full MLPerf® Training V6.0 benchmark results have been submitted.

The details: - ClusterReady™ uses inference and training workloads alongside XPerf’s proprietary test suite. - The system stresses GPUs, memory, interconnects, network fabric and thermal systems under conditions meant to mirror live production. - XPerf says ClusterReady™ measures tokens per second output under sustained load. - XPerf says the validation report gives operators a revenue-linked performance baseline for each rack. - When issues are found, XPerf’s algorithms identify the root cause and prescribe remediation or optimization steps. - XPerf says the tool reduces manual engineering effort. - XPerf says network optimization delivered a 10× improvement in multi-node communication bandwidth across the full 4-node rack. - The company says that enabled 95.8% scaling efficiency on Llama 3.1 8B pre-training. - XPerf says ClusterReady™ is commercially available for supported environments. - XPerf directs users to the ClusterReady™ demo page for more information.

Between the lines: - The pitch is about proving not just that a rack works, but that it performs well enough to matter economically. - By tying validation to sustained tokens-per-second output, XPerf is trying to make rack acceptance more directly linked to real-world AI throughput. - The 20% benchmark gain and 10× bandwidth improvement are based on internal testing in a specific MiTAC configuration, so they are not automatically transferable to other deployments. - XPerf’s forward-looking language signals that broader market results could vary by software version, workload and system setup.

What's next: - XPerf says it has already submitted full MLPerf Training V6.0 results. - The company is promoting ClusterReady™ for rapid deployment in supported environments. - XPerf is also positioning the product as part of a broader lifecycle that includes its OpsPilot™ autonomous operations software.

The bottom line: - XPerf wants AI infrastructure buyers to treat validation as a performance and revenue decision, not just an integration check.