Intel’s newest flagship workstation graphics cards, the Arc Pro B65 and Arc Pro B70, launched around two weeks ago. Yet despite the buzz around Intel’s “Big Battlemage” workstation lineup, real-world performance details have been difficult to find. Review coverage is still sparse, and availability appears limited—similar to what was seen with last year’s Arc Pro B60, where supply issues meant many creators and professionals barely had a chance to evaluate the card firsthand.
These Arc Pro models are designed specifically for professional workloads and AI tasks, focusing on higher memory capacity and stronger GPU horsepower than the Arc Pro B50 and B60. That makes the Arc Pro B70 particularly interesting for users building workstations for inference, content creation pipelines, visualization, and other compute-heavy jobs where VRAM and efficiency matter.
So far, most performance claims have come from Intel’s own benchmarks, which suggest the Arc Pro B70 can outperform the NVIDIA RTX PRO 4000 in certain workloads. But independent testing is what many workstation buyers wait for—and one of the few detailed third-party looks comes from Hardware Luxx, which managed to obtain four Arc Pro B70 cards. That access allowed not only single-card testing, but also a rare look at multi-GPU scaling with a quad-GPU configuration.
In AI-focused benchmarks, the Arc Pro B70 reportedly lands around 30% to 40% faster than the Arc Pro B50 in Geekbench AI. In several tests, it also goes toe-to-toe with AMD’s Radeon Pro W7900, which is notable because the Arc Pro B70 is said to draw considerably less power than AMD’s higher-end workstation options like the Radeon Pro W7800 and W7900.
Power efficiency is one of the most eye-catching takeaways. Under full load in inference workloads, the Arc Pro B70 was observed at roughly 180W per card. That means a dual-card setup scales in a predictable way at around double the power draw, and the same pattern continues with four cards installed. In the quad-GPU configuration, total power consumption was reported at around 720W—still below what you might expect if each card were constantly hitting its official rated limits.
Intel’s Arc Pro B70 reference model is officially rated at 230W, but testing indicates it may not reach that ceiling during inference tasks, topping out closer to 180W in the workloads measured. For professionals, that gap can be important: lower real-world power draw can translate into easier thermal management, potentially quieter systems, and more flexible workstation configurations—especially when scaling up to multiple GPUs.
On the specification side, the Arc Pro B70 is built around the BMG-G31 die and includes 32 Xe cores. It comes equipped with 32GB of GDDR6 memory running on a 256-bit memory bus, along with a full PCIe 5.0 x16 interface. That combination targets the needs of AI inference and pro workflows where large datasets and memory-heavy models can quickly overwhelm lower-VRAM GPUs.
While broader availability and more independent benchmarks will ultimately determine how competitive the Arc Pro B70 really is in the workstation GPU market, early signs point to a capable AI-focused card with promising efficiency—especially for users interested in multi-GPU setups and performance per watt.
