GPUs & Macs › Intel Arc A750
Intel Arc A750: local LLM speed
Llama 2 7B (Q4_0) generates 44.0 tokens/s on the Intel Arc A750 (Vulkan), and processes prompts at 303 tokens/s.
Measured results
| Model | Generate tok/s | Prompt tok/s | Quant | Backend | Test | Source |
|---|---|---|---|---|---|---|
| Llama 2 7B | 44.0 | 303 | Q4_0 | Vulkan FA on | tg128 | llama.cpp #10879 details |
| Llama 2 7B | 42.7 | 1,076 | Q4_0 | Vulkan FA off | tg128 | llama.cpp #10879 details |
What fits on the Intel Arc A750
Estimates for this chip are scaled from a measured run: Llama 2 7B Q4_0 at 44.0 tok/s (Vulkan, source). Fit assumes the whole model in GPU memory with ~8K context.
| Model | Best quant that fits | File | Expected tok/s |
|---|---|---|---|
| gemma-4-12B-it | IQ4_XS | 6.4 GB | 26.3 estimate |
| Qwen3.5-9B | Q5_K_S | 6.4 GB | 26.4 estimate |
| Qwen3.5-4B | Q8_0 | 4.5 GB | 37.5 estimate |
| MiMo-V2.6-Distill-Qwen-9B | Q4_K_M | 5.8 GB | 28.8 estimate |
| gemma-4-E4B-it | Q5_K_M | 5.5 GB | 30.6 estimate |
| Qwen3-4B | Q8_0 | 4.3 GB | 39.2 estimate |
| gemma-3-1b-it | Q8_0 | 1.1 GB | 140 estimate |
| gemma-4-E2B-it | Q8_0 | 5.0 GB | 33.8 estimate |
| DeepSeek-R1-0528-Qwen3-8B | Q5_K_M | 5.9 GB | 28.7 estimate |
| Llama-3.2-3B-Instruct | Q8_0 | 3.4 GB | 49.1 estimate |
| Phi-4-mini-instruct | Q8_0 | 4.1 GB | 41.1 estimate |
| gemma-3-4b-it | Q8_0 | 4.1 GB | 40.7 estimate |
| Qwen3-14B | UD-IQ3_XXS | 6.0 GB | 27.9 estimate |
| gemma-3-12b-it | Q3_K_M | 6.0 GB | 28.0 estimate |
Similar hardware
Intel Arc B570 · Intel Arc B580 · Intel Arc Pro B70 · Intel Arc A770