Intel has launched the Arc Pro B70 and B65, its "Big Battlemage" workstation GPUs featuring 32GB of GDDR6 memory and up to 367 TOPS of AI compute, starting at $949 for the B70.

What Happened

Announced at Intel Pro Day 2026, the Arc Pro B70 packs the full BMG-G31 GPU with 32 Xe2-HPG cores, 256 XMX AI engines, and 32 ray tracing units. The 32GB of GDDR6 memory runs across a 256-bit bus at 19 Gbps, delivering 608 GB/s of bandwidth. Tom's Hardware reports the B70 is available now from partners including ARKN, ASRock, Gunnir, Maxsun, and Sparkle.

The Arc Pro B65 uses 20 Xe2-HPG cores and 160 XMX engines for 197 TOPS while keeping the full 32GB memory and 608 GB/s bandwidth. XDA notes the B65 ships in mid-April at a lower price point yet to be announced.

Why It Matters

For creators running AI models locally, VRAM is the bottleneck that determines which models you can load. Most consumer GPUs top out at 16GB, which is not enough for the latest image and video generation models at full quality. The Arc Pro B70's 32GB at $949 undercuts NVIDIA RTX Pro workstation GPUs that charge significantly more for equivalent memory.

The B65 is particularly interesting for inference-heavy workflows. It keeps the same 32GB memory pool as the B70 at a lower price, making it a strong option for creators whose work is memory-bound rather than compute-bound. Running local AI workflows with large models like Flux or video generation pipelines could become accessible on a sub-$1,000 GPU budget.

Key Details

  • Arc Pro B70: 32 Xe2-HPG cores, 256 XMX engines, 367 INT8 TOPS, 32GB GDDR6, $949, available now
  • Arc Pro B65: 20 Xe2-HPG cores, 160 XMX engines, 197 INT8 TOPS, 32GB GDDR6, mid-April launch
  • Both cards deliver 608 GB/s memory bandwidth
  • Targeted at content creation, engineering, and AI inferencing
  • Partners: ARKN, ASRock, Gunnir, Maxsun, Sparkle

What to Do Next

If you have been waiting for affordable high-VRAM GPUs for local AI work, the B70 is worth evaluating against NVIDIA's RTX Pro lineup. Check software compatibility first. Intel's driver support for AI frameworks has improved but still lags behind NVIDIA CUDA and AMD ROCm in some workflows. The B65 at a lower price could be the better value if your models are memory-limited rather than compute-limited.