GPUs & Macs › Intel Arc B580
Intel Arc B580: local LLM speed
Llama 2 7B (Q4_0) generates 72.0 tokens/s on the Intel Arc B580 (Vulkan), and processes prompts at 419 tokens/s.
Measured results
| Model | Generate tok/s | Prompt tok/s | Quant | Backend | Test | Source |
|---|---|---|---|---|---|---|
| Llama 2 7B | 72.0 | 419 | Q4_0 | Vulkan FA on | tg128 | llama.cpp #10879 details |
| Llama 2 7B | 70.1 | 621 | Q4_0 | Vulkan FA off | tg128 | llama.cpp #10879 details |
What fits on the Intel Arc B580
Estimates for this chip are scaled from a measured run: Llama 2 7B Q4_0 at 72.0 tok/s (Vulkan, source). Fit assumes the whole model in GPU memory with ~8K context.
| Model | Best quant that fits | File | Expected tok/s |
|---|---|---|---|
| Qwen3.8-27B | UD-IQ2_XXS | 7.3 GB | 37.9 estimate |
| gemma-4-12B-it | Q6_K | 9.8 GB | 28.1 estimate |
| Qwen3.5-9B | Q8_0 | 9.5 GB | 28.9 estimate |
| Qwen3.5-4B | Q8_0 | 4.5 GB | 61.4 estimate |
| gemma-4-26B-A4B-it | UD-IQ2_M | 10.0 GB | 89.3 estimate |
| MiMo-V2.6-Distill-Qwen-9B | Q8_0 | 9.5 GB | 28.8 estimate |
| gemma-4-E4B-it | Q8_0 | 8.2 GB | 33.6 estimate |
| Qwen3-4B | Q8_0 | 4.3 GB | 64.3 estimate |
| gemma-4-31B-it | UD-IQ2_XXS | 8.5 GB | 32.2 estimate |
| gemma-4-12B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 6.7 GB | 41.0 estimate |
| gemma-3-1b-it | Q8_0 | 1.1 GB | 229 estimate |
| gemma-4-E2B-it | Q8_0 | 5.0 GB | 55.4 estimate |
| DeepSeek-R1-0528-Qwen3-8B | Q8_0 | 8.7 GB | 31.6 estimate |
| Llama-3.2-3B-Instruct | Q8_0 | 3.4 GB | 80.4 estimate |
Similar hardware
Intel Arc B570 · Intel Arc A770 · Intel Arc A750 · Intel Arc Pro B70