# Arc B580

> OnPrem World · https://onpremworld.com/en/hardware/intel-arc-b580 · Last updated: 2026-10-16 · CC BY 4.0

The cheapest card here at launch ($249), with 12 GB of VRAM for 8–14B models through llama.cpp's SYCL or Vulkan backends.

## Specifications

|  |  |
|---|---|
| Memory | 12 GB GDDR6 |
| Memory bandwidth | 456 GB/s |
| Power | 190 W |
| Launch | 2024-12 |
| Launch price | US$249 |
| Architecture | Battlemage (Xe2) |
| Software | oneAPI (SYCL) / Vulkan |

## What it can run (12 GB, 8K)

Estimates, not benchmarks. See the methodology.

| Model | Parameters | Best quant | Total | tok/s (est.) |
|---|---|---|---|---|
| Phi-4 (14B) | 14B | Q5_K_M | 11.8 GiB | 30 |
| Gemma 3 12B IT | 12B | Q6_K | 11.0 GiB | 30 |
| Gemma 4 12B Unified IT | 11.95B | Q6_K | 10.5 GiB | 30 |
| Qwen3.5-9B | 9B | Q8_0 | 10.1 GiB | 31 |
| Qwen3-8B | 8.2B | Q8_0 | 10.1 GiB | 34 |
| DeepSeek-R1-0528-Qwen3-8B | 8.19B | Q8_0 | 10.1 GiB | 34 |
| Llama 3.1 8B Instruct | 8B | Q8_0 | 9.8 GiB | 35 |
| Gemma 4 E4B IT | 8B | Q8_0 | 9.1 GiB | 35 |
| Llama 3.2 3B Instruct | 3.21B | FP16 | 7.7 GiB | 45 |
| Kimi K2.6 | 1000B | Doesn't fit | – | – |
| DeepSeek-R1-0528 (671B) | 671B | Doesn't fit | – | – |
| DeepSeek-V4-Flash | 284B | Doesn't fit | – | – |
| Qwen3.8-Flash-Next | 180B | Doesn't fit | – | – |
| Qwen3.5-122B-A10B | 122B | Doesn't fit | – | – |
| Mistral Small 4 119B (2603) | 119B | Doesn't fit | – | – |
| gpt-oss-120b | 117B | Doesn't fit | – | – |
| Llama 4 Scout 17B-16E Instruct | 109B | Doesn't fit | – | – |
| GLM-4.5-Air | 106B | Doesn't fit | – | – |
| Qwen3-Coder-Next (80B-A3B) | 80B | Doesn't fit | – | – |
| Llama 3.3 70B Instruct | 70B | Doesn't fit | – | – |
| Qwen3.5-35B-A3B | 35B | Doesn't fit | – | – |
| Qwen3.6-35B-A3B | 35B | Doesn't fit | – | – |
| Qwen3-32B | 32.8B | Doesn't fit | – | – |
| DeepSeek-R1-Distill-Qwen-32B | 32.76B | Doesn't fit | – | – |
| Gemma 4 31B IT | 30.7B | Doesn't fit | – | – |
| Qwen3-30B-A3B (2507) | 30.5B | Doesn't fit | – | – |
| Qwen3-Coder-30B-A3B-Instruct | 30.5B | Doesn't fit | – | – |
| GLM-4.7-Flash (30B-A3B) | 30B | Doesn't fit | – | – |
| Qwen3.5-27B | 27B | Doesn't fit | – | – |
| Qwen3.6-27B | 27B | Doesn't fit | – | – |
| Qwen3.8-27B | 27B | Doesn't fit | – | – |
| Gemma 3 27B IT | 27B | Doesn't fit | – | – |
| Gemma 4 26B A4B IT | 25.2B | Doesn't fit | – | – |
| Mistral Small 3.2 24B Instruct (2506) | 24B | Doesn't fit | – | – |
| Magistral Small 1.2 (2509) | 24B | Doesn't fit | – | – |
| Devstral Small 2 24B Instruct (2512) | 24B | Doesn't fit | – | – |
| gpt-oss-20b | 21B | Doesn't fit | – | – |

## Prices

Latest price at each shop, as listed on the day we checked.

| Region | Shop | Listing | Price | Per GB | Checked |
|---|---|---|---|---|---|
| US | Newegg | ASRock Challenger Arc B580 12GB GDDR6 PCI Express 4.0 x8 ATX Graphics Card B580 CL 12GO | [US$329.99](https://www.newegg.com/asrock-challenger-b580-cl-12go-arc-b580-12gb-graphics-card-double-fans/p/N82E16814930131) | US$27/GB | 2026-10-14 |

US value: about 58.2 tok/s on Llama 3.1 8B Instruct (Q4_K_M, 8K), or 17.6 tok/s per US$100.

All prices: https://onpremworld.com/en/prices.md

## Frequently asked questions

### Which LLMs can the Arc B580 run?

With 12 GB and an 8K context, the largest models that fit include: Phi-4 (14B), Gemma 3 12B IT, Gemma 4 12B Unified IT, Qwen3.5-9B, Qwen3-8B. See the full table below.

### How fast is the Arc B580 for local LLMs?

Generation speed is mostly set by memory bandwidth (456 GB/s). For Llama 3.1 8B Instruct at Q8_0, expect roughly 35 tokens/s; bigger models are slower.

### Can the Arc B580 run 70B models?

Llama 3.3 70B Instruct does not fit on the Arc B580 (12 GB), even at Q4_K_M with an 8K context.

## Sources

- https://www.intel.com/content/www/us/en/products/sku/241598/intel-arc-b580-graphics/specifications.html
- https://www.intel.com/content/www/us/en/newsroom/news/intel-launches-arc-b-series-graphics-cards.html
- https://github.com/ggml-org/llama.cpp/blob/master/docs/backend/SYCL.md