Intel Arc Pro B60 24GB
No editorial image yet — generic vendor mark shown. Credentials in spec table below.
Intel's workstation card explicitly marketed for low-cost local LLM inference. 24GB GDDR6, 456 GB/s, ~197 TOPS, ~$599. Board partners ship 48GB dual-GPU variants; Intel's 'Project Battlematrix' scales to 8x = 192GB for 70B+ serving.
Sub-scores sum to 371 / 1000. Headline = 371 × 0.70 (Estimated-confidence discount) = 260. This is an algorithmic performance-tier score — distinct from, and often lower than, the editorial “Our verdict” below, which weighs value and real-world fit (especially for hardware we haven’t measured yet). How scoring works →
Extrapolated from 456 GB/s bandwidth — 36.5 tok/s estimated. No measured benchmarks yet.
Plain-English: Workable at 32B, comfortable at 14B and below — coding agent feels deliberate.
Verdicts extrapolated from catalog VRAM + bandwidth + ecosystem flags. Hover any chip for the rationale. Want measured numbers? Submit your own run with runlocalai-bench --submit.
What it does well
The Arc Pro B60 is the most interesting value play in local-AI hardware: 24GB of VRAM for ~$599 — roughly half the per-GB cost of NVIDIA's 24GB+ options. Intel built it specifically for private local inference, and dual-GPU board-partner cards (48GB on one slot) plus the Battlematrix platform (up to 8 cards = 192GB) make it a genuinely cheap path to serving 70B+ models on-prem. With vLLM and Intel's LLM Scaler software, throughput on supported models is competitive for the price.
Where it struggles
The software ecosystem is the gamble. There's no CUDA — you're on Intel's oneAPI/IPEX-LLM, SYCL, and Vulkan paths, which have improved a lot but still trail NVIDIA on day-one model support, quantization-format breadth, and the long tail of community repos that assume CUDA. Expect to do more setup work and hit occasional unsupported-op walls. Raw compute is modest, so prefill on long prompts is slower than a comparable NVIDIA workstation card.
Bottom line
The best $/VRAM in local AI right now and a real option for budget on-prem 70B serving — if you're comfortable living in Intel's software stack rather than CUDA's. For maximum compatibility and least friction, NVIDIA still wins; for cheap VRAM at scale, the B60 is unmatched.
Overview
Intel's workstation card explicitly marketed for low-cost local LLM inference. 24GB GDDR6, 456 GB/s, ~197 TOPS, ~$599. Board partners ship 48GB dual-GPU variants; Intel's 'Project Battlematrix' scales to 8x = 192GB for 70B+ serving.
Search-fallback link — editorial hasn't yet curated a retailer URL for this card.
Some links above are affiliate links. We may earn a commission at no extra cost to you. How we make money.
Specs
| VRAM | 24 GB |
| Power draw (peak) | 200 W |
| Released | 2025 |
| MSRP | $599 |
| Backends | Vulkan |
Models that fit
Open-weight models small enough to run on Intel Arc Pro B60 24GB with usable context.
Frequently asked
What models can Intel Arc Pro B60 24GB run?
Does Intel Arc Pro B60 24GB support CUDA?
Where next?
Reviewed by RunLocalAI Editorial. See our editorial policy for how we research and verify hardware specifications.