Intel Xeon gains 2.4x in MLPerf v6.1 via software

MLPerf Inference is a widely watched set of benchmarks run by MLCommons that lets companies compare how fast different hardware handles AI tasks. In the v6.0 round, Intel's submissions focused on a narrower set of Xeon and Arc Pro configurations. For v6.1, Intel expanded its Xeon participation from two benchmarked SKUs to five, increasing CPU inference results from 24 to 35, and its Arc Pro B70 submissions now cover more models including Llama 2 70B and gpt-oss-120B.

The company also pointed to broader ecosystem involvement, with Oracle making its first Intel-based submission and Red Hat delivering its first Xeon CPU inference submission. Quanta Cloud Technology and Supermicro provided the first partner submissions using Intel Arc Pro B70 GPUs. Intel says its Xeon improvements are being upstreamed into widely used AI frameworks so customers can benefit on deployed infrastructure, and its Arc Pro B70 optimization work is intended to advance software for future Intel GPU products.

The stakes for Intel's argument rest on whether these software gains translate into real cost savings for businesses running AI inference. That likely depends on how closely a customer's setup matches the benchmarked configurations and whether they adopt the updated software. If the improvements hold in typical enterprise environments, they could help delay hardware replacement cycles, a meaningful consideration for IT teams managing tight budgets.