Intel has begun shipping its Arc Pro B50 16GB professional GPU to reviewers, showcasing the compute and AI performance available for $349. The card was put through a variety of workstation-focused workflows, from design and engineering to image editing and AI.
As a reminder, the Intel Arc Pro B50 is based on BMG-G21 silicon featuring 16 Xe2 cores and 16GB of 14Gb/s GDDR6 memory running on a 128-bit bus. This results in 224GB/s of memory bandwidth, which should be enough to keep the GPU well fed. Most importantly, the card is quite compact, taking only two slots in its low-profile format, all while requiring just 70W of power, removing the need for any power cables.
Unlike the Arc Pro B60, which wasn’t sampled to reviewers for some reason, B50 was received by at least three outlets, including HardwareLuxx, Phoronix, and Igor’s LAB. Each put the card through their preferred software testing suite, comprised of tools such as Procyon AI, SPECviewperf, Photoshop, AutoCAD, Blender, and 3DMark.
Starting with HardwareLuxx, the outlet found that while the Arc Pro B50 delivered better AI performance than the Nvidia RTX A1000, both in metrics such as tokens per second (TPS) and Time to First Token (TTFT), it couldn’t steer ahead in compute benchmarks. Though still nice considering both cards retail within $15 of each other, I note the Nvidia chip is quite old, dating back to May 2020, showing the indirect weakness of Intel’s 2025 architecture. Of course, things are unlikely to get better against Nvidia’s RTX Pro 2000 Blackwell solution with its 16GB of memory.
Phoronix, on the other hand, found that the Arc Pro B50 was quite potent in synthetic and real-world tests that called on OpenGL and Vulkan. There, the Arc Pro was 1.47x faster than the RTX A1000, which is slightly below Intel’s internal claims of 1.6x. Even so, the B50 offers solid value, especially for those looking for a cheap AI-capable GPU, where its 16GB of VRAM is handy.
Lastly, Igor’s LAB found the Arc B50 great in Photoshop, After Effects, and Premiere Pro, beating even the more expensive AMD Pro W7500. But that’s not all, the biggest gains were found in AI tasks such as Computer Vision and Stable Diffusion when using the Intel-optimised OpenVINO toolkit.
It must be said that even using Microsoft ONNX, Pro B50 held its ground against the RTX A1000 and the Pro W7500, just not to the same level as with OpenVINO. Perusing the results, it’s also worth noting that the RTX failed some of these tests due to its insufficient 8GB memory. This shows that 16GB of VRAM is mandatory in certain situations, relegating 8GB cards to less complex models.
Overall, these are some positive results, especially if you plan on getting a B50 for AI workloads. In case you were interested, the card is available both as a standalone purchase or as part of prebuilt systems. So, there is something for everyone.
