Intel’s Project Battlematrix Featuring Arc Pro B60 GPUs Achieves 4x Performance Per Dollar Compared to NVIDIA in MLPerf v5.1 Benchmarks

Intel has released new benchmark results for its Project Battlematrix workstation, featuring Arc Pro B60 GPUs, as part of the MLPerf Inference v5.1 evaluations.

Performance Insights: Intel Project Battlematrix with Arc Pro B60 GPUs and Xeon 6 CPUs

Press Release: In a significant development, MLCommons has unveiled the latest MLPerf Inference v5.1 benchmarks, showcasing Intel’s advanced GPU systems. These systems utilize Intel Xeon processors paired with P-cores and Arc Pro B60 GPUs, collectively referred to as Project Battlematrix.

The benchmarks indicate that the Intel Arc Pro B60 achieves up to 1.25 times better performance per dollar in the Llama 8B benchmark, and demonstrates an impressive up to fourfold performance advantage over NVIDIA’s RTX Pro 6000 and L40S. This performance highlight marks a significant step in making advanced AI inference solutions accessible primarily through all-Intel hardware.

The Significance of This Development

For professionals requiring robust AI inference platforms, the options have been notably limited. The need for a solution that could deliver excellent inference performance while ensuring data privacy—and avoiding the hefty subscription fees associated with proprietary AI models—has been a growing concern. The Project Battlematrix initiative aims to fill this gap by providing a comprehensive solution that can deploy large language models (LLMs) effectively.

Intel’s new GPU systems are crafted not only to address modern AI tasks but also to provide an all-inclusive inference platform, fusing reliable hardware with validated software frameworks. The focus on user experience is evident, as these systems come equipped with a new containerized solution optimized for Linux environments, facilitating exceptional inference performance through multi-GPU scaling and PCIe point-to-point data transfers.

The Essential Role of CPUs in AI Systems

CPUs remain central to the architecture of AI systems, serving as the critical orchestrator for processing, transmission, and overall system management. Over the past four years, Intel has consistently improved its CPU performance in AI applications. These advancements have positioned Intel Xeon processors as the preferred choice for effectively managing AI workloads within GPU-powered systems.

Intel’s commitment to excellence is further highlighted by its status as the only vendor submitting server CPU results to MLPerf, showcasing its dedication to enhancing AI inference capabilities across both computational and accelerator platforms. Notably, the introduction of the Intel Xeon 6 processor with P-cores has yielded a 1.9 times performance enhancement generation over generation in the latest MLPerf Inference v5.1 benchmarks.

Source & Images