For enterprise buyers looking to maximize performance per dollar on AI workloads, Intel's new Arc Pro GPUs present a compelling alternative to established players.
The Arc Pro B70 stands out as the most powerful discrete GPU based on Intel's Xe2 'Battlemage' architecture. With 32 Xe cores, a 256-bit GDDR6 memory interface, and 32 GB of VRAM, it delivers up to 367 TOPS of INT8 compute performance. This configuration addresses a long-standing gap in Intel's lineup, where the Arc B580 gaming GPU had been limited to 24 Xe cores and a 192-bit interface.
The B70 is designed for advanced AI compute tasks, including local inferencing and multi-GPU configurations, with support for PCI-Express 5.0 x16 and four DisplayPort 2.1 outputs. Its power draw ranges from 160 W to 290 W, depending on the implementation.
Complementing the B70 is the Arc Pro B65, which shares a similar core configuration with 24 Xe cores but introduces key improvements. It features 32 GB of GDDR6 memory across a 192-bit interface and utilizes PCI-Express 5.0 x16, unlike the Arc B580's PCIe 5.0 x8. This variant is positioned as a more balanced option for workstations, with a typical board power of 200 W.
Both GPUs support Intel's oneAPI, OpenCL 3.0, and OpenVINO compute APIs, along with hardware-accelerated media encoding for AV1, HEVC, VP9, and H.265. Certified drivers will be available for Windows 11, Windows 10, and Linux.
What remains unclear is the pricing strategy for these GPUs. While the B70's specifications suggest a premium positioning, Intel has not yet disclosed exact price points. Enterprise buyers should weigh the performance gains against the potential cost implications, particularly in the context of current memory market conditions.
The introduction of these GPUs underscores Intel's commitment to expanding its discrete GPU footprint in AI and professional visualization markets. For enterprises evaluating their options, the Arc Pro B70 and B65 offer significant compute capabilities that could influence future investment decisions.
