IN Brief:
- Intel has published new MLPerf Inference v6.0 results built around Xeon 6 CPUs and Arc Pro B-Series GPUs.
- The company says a four-GPU Arc Pro B70 or B65 system provides 128GB of VRAM for larger model execution.
- Intel is positioning the stack around multi-GPU scaling, open software, and inference deployment across workstation, edge, and datacentre platforms.
Intel has published new MLPerf Inference v6.0 benchmark results centred on Xeon 6 processors and Arc Pro B-Series GPUs, setting out its latest case for open, scalable AI inference platforms across workstation, edge, and datacentre deployments. The company said the results cover four benchmark submissions for GPU-based systems using Xeon 6 host processors alongside Arc Pro B70 graphics.
Intel said a four-GPU Arc Pro B70 or B65 configuration delivers 128GB of VRAM, allowing execution of models up to 120 billion parameters with high concurrency. It also said the Arc Pro B70 delivers up to 1.8x higher inference performance than the Arc Pro B60 in the cited configuration, while software changes in its containerised stack produced up to 1.18x higher gains on the same Arc Pro B60 hardware compared with the earlier MLPerf v5.1 round.
The company is also leaning on platform features beyond raw accelerator throughput. Intel said the Arc Pro B-Series stack supports ECC memory, SR-IOV virtualisation, telemetry, remote firmware updates, and PCIe peer-to-peer transfers, while Xeon 6 remains central to host-side memory management and workload orchestration. Intel added that more than half of MLPerf Inference v6.0 submissions used Xeon processors as host CPUs, and said Xeon 6 with performance cores delivered up to 1.9x improvement in MLPerf Inference v5.1 over the previous generation.
