Intel launches Arc Pro B70 and B65 with 32GB GDDR6 featuring “Big Battlemage” GPU

Intel has introduced two new Arc Pro workstation GPUs, the Arc Pro B70 and Arc Pro B65. Both cards use the larger BMG-G31 GPU and both come with 32GB of GDDR6 memory. Intel is positioning them for local AI inference, agent workloads, software development, and professional graphics applications.

Arc Pro B70 is the top model in the new launch. Intel lists 32 Xe cores, 256 XMX engines, up to 367 peak TOPS, and 608 GB/s memory bandwidth. Board power is listed at 160W to 290W, while Intel’s own branded card is rated at 230W. The company says the B70 will be available starting March 25 from Intel and partners including ASRock, Gunnir, MAXSUN, Sparkle, Senao, Lanner, Onix, and others, with pricing starting at $949.

Source: Intel

Arc Pro B65 uses a cut-down BMG-G31 configuration with 20 Xe cores and 160 XMX engines. It still carries 32GB of GDDR6 memory and is rated for up to 197 peak TOPS. Board power is listed at 200W. Intel says this model will launch in mid-April through add-in-board partners, and it will be priced below the Arc Pro B70. Intel has not announced pricing for B65 yet.

Source: Intel

Source: Intel

Intel’s own material puts most of the focus on Linux-based AI inference. In those slides, the company compares Arc Pro B70 against NVIDIA RTX Pro 4000 and claims up to 2.2x larger context windows, up to 85% higher token throughput in multi-user workloads, up to 6.2x faster time to first token, and up to 2x better token-per-dollar in selected tests. Intel also claims up to 69% higher performance for professional applications versus Arc Pro B60 in SPECviewperf 15.

1 / 11

1 / 11

This launch confirms that Intel’s bigger Battlemage GPU is arriving first as a workstation and AI product, not as a gaming card. For now, Arc Pro B70 looks like the main part, bringing 32GB memory, multi-GPU support, and a much higher AI throughput target than the existing B60. Arc Pro B65 keeps the same 32GB capacity, but in a lower tier aimed at buyers who need memory headroom more than peak compute.

Source: Intel PR

Follow us on Google News Set as Preferred Source

Share