Intel Project Battlematrix Challenges NVIDIA with Arc Pro B60 GPUs

Intel is stepping up its presence in the AI inference market with fresh benchmark results for its Project Battlematrix workstation powered by Arc Pro B60 GPUs and Intel Xeon processors.

According to the newly released MLPerf Inference v5.1 benchmarks, the all-Intel setup delivers compelling performance gains and significant cost efficiency when compared with competing NVIDIA solutions.

The headline figure comes from tests on the Llama 8B model, where Intel’s Arc Pro B60 demonstrated up to 1.25x better performance per dollar versus NVIDIA’s RTX Pro 6000 and up to 4x better than the NVIDIA L40S. This sharp advantage underscores Intel’s positioning of Battlematrix as an accessible, cost-effective alternative for professionals needing high-performance AI inference without vendor lock-in or steep subscription fees tied to proprietary stacks.

For years, AI developers and enterprises faced a narrow range of hardware options. The landscape has largely been dominated by NVIDIA, leaving many frustrated with high entry costs and a fragmented ecosystem. Intel’s Battlematrix aims to change this dynamic by offering a validated full-stack platform that blends CPUs and GPUs into a containerized Linux solution. The system supports multi-GPU scaling, PCIe peer-to-peer data transfers, ECC memory, SR-IOV virtualization, remote firmware updates, and detailed telemetry – features traditionally reserved for high-end enterprise deployments.

Equally important is the role of CPUs in AI inference. While GPUs accelerate the math-intensive cores of large language models, CPUs remain the orchestration hub, managing preprocessing, data transfer, and system coordination. Intel has invested heavily in improving CPU inference performance, with Xeon 6 featuring P-cores showing nearly 1.9x generation-over-generation gains in MLPerf v5.1. Intel also remains the only vendor consistently submitting server CPU results to MLPerf, signaling its broader commitment to hybrid CPU+GPU acceleration.

Still, challenges remain. Some professionals note that while the benchmarks look promising on paper, availability of Arc Pro B60 cards in the market remains scarce, and Intel’s software ecosystem is not yet as mature or developer-friendly as NVIDIA’s CUDA-dominated environment. Despite these issues, the Battlematrix project suggests Intel is serious about reshaping the inference workstation space, offering a more open and competitive platform for AI builders, researchers, and edge deployment scenarios.

If Intel can deliver these systems at scale with strong driver and software support, it could provide much-needed competition in a market where developers have long sought affordable, powerful, and enterprise-ready alternatives.