Intel Arc B580 for local LLMs
Written by Jakub Rusinowski · Last updated
With 12 GB at 456 GB/s, the Intel Arc B580 runs 69 catalogued models at Q4_K_M with 8K context. The largest that fits is Gemma 3 12B Instruct (~11.1 GB), and the top pick is Cosmos 3 Nano at 24–50 tok/s.
Surprisingly capable for AI at $249. 12 GB VRAM with OpenVINO/IPEX-LLM acceleration. Works with Ollama on Windows and Linux. Best Intel GPU for local AI in 2024–2025.
Models that run on the Intel Arc B580
Q4_K_M, 8K context, 12 GB usable. Ranked by quality and speed.
| Model | Memory · marker = 12 GB | VRAM | Speed |
|---|---|---|---|
| Cosmos 3 Nano Cosmos 3 | 10.5 GB | 24–50 tok/s | |
| Ministral 3 14B Ministral 3 | 9.3 GB | 16–34 tok/s | |
| DeepSeek R1 Distill Qwen 14B DeepSeek R1 | 10.6 GB | 16–34 tok/s | |
| Gemma 4 12B (Unified) Gemma 4 | 8 GB | 18–38 tok/s | |
| Mistral NeMo 12B Mistral Family | 9.4 GB | 18–38 tok/s | |
| Bielik PL 11B v3.0 Instruct Bielik | 7.4 GB | 20–41 tok/s | |
| Llama 3.2 Vision 11B Llama 3.2 Vision | 8.5 GB | 20–42 tok/s | |
| Llama 3.2 11B Vision Instruct Llama 3.2 Family | 8.5 GB | 20–42 tok/s | |
| Falcon 3 10B Instruct Falcon 3 | 8.4 GB | 21–43 tok/s | |
| Qwen 3.5 9B Qwen 3.5 | 6.5 GB | 26–54 tok/s | |
| GLM-4 9B GLM-4.7 / GLM-Z1 | 6.2 GB | 23–48 tok/s | |
| K2 Horizon 7B K2 Horizon | 7.4 GB | 23–48 tok/s |
Buy it or rent the same memory
Buy the card, or rent a GPU with the same memory by the hour to try models first.
or compare on Vast.ai from $0.35/hr (typical low · varies)
As an Amazon Associate we earn from qualifying purchases. Cloud GPU links are referral links — we may earn a commission at no extra cost to you.
Speed vs other GPUs
Llama 3.1 8B, Q4_K_M. Estimated ranges. How this is calculated
Specifications
Specs last updated 2026-07-12.
- Memory
- 12 GB
- Memory bandwidth
- 456 GB/s
- Architecture
- Xe2 Battlemage BMG-G21
- Series
- Arc Battlemage
- Board power
- 190 W
- Release year
- 2024
- Launch price
- $249
- Usable for models
- 12 GB
Similar GPUs
Frequently asked questions
Can the Intel Arc B580 run local LLMs?
Yes. With 12 GB (12 GB usable by a model) it runs 69 of the catalogued models at Q4_K_M with 8K context; the largest is Gemma 3 12B Instruct, needing about 11.1 GB.
How fast is the Intel Arc B580 for AI inference?
It is estimated to run Llama 3.1 8B at 28–57 tok/s at Q4_K_M. Llama 3.3 70B does not fit: it needs about 44 GB against 12 GB usable. These are modelled estimates from memory bandwidth, not measurements; the methodology page shows the formula.
What LLMs can I run on 12 GB?
Among the best that fit: Cosmos 3 Nano, Ministral 3 14B, DeepSeek R1 Distill Qwen 14B, Gemma 4 12B (Unified), Mistral NeMo 12B. The quickest start is Ollama: ollama run llama3.1:8b.
Keep going
Can I run it on the Intel Arc B580?
More for this card
Tools