Home › Models › Qwen3.5 35B-A3B
Qwen3.5 35B-A3B: VRAM needs and speed by GPU
At Q4_K_M, Qwen3.5 35B-A3B needs about 23.7 GB of VRAM. We have 1 measured speed results for it.
34.66B parameters (mixture-of-experts, ~3.0B active) · license: apache-2.0 · GGUF files on Hugging Face
Measured speeds
| Chip | Generation tok/s | Prompt tok/s | Quant | Backend | Test | Source |
|---|---|---|---|---|---|---|
| Intel Arc Pro B70 | 54.5 | 618 | Q4_K_M | SYCL FA on | tg128 | PMZFX intel-arc-pro-b70-benchmarks (MIT) details |
Memory needed per quantization
| Quant | File size | Needs (estimate) | Smallest card it fits |
|---|---|---|---|
| UD-IQ2_XXS | 10.7 GB | 12.3 GB | 16 GB cards |
| UD-IQ2_M | 11.4 GB | 13.1 GB | 16 GB cards |
| UD-Q2_K_XL | 12.2 GB | 13.8 GB | 16 GB cards |
| UD-IQ3_XXS | 13.1 GB | 14.8 GB | 16 GB cards |
| UD-IQ3_S | 13.6 GB | 15.3 GB | 16 GB cards |
| Q3_K_S | 15.3 GB | 16.9 GB | 24 GB cards |
| Q3_K_M | 16.4 GB | 18.0 GB | 24 GB cards |
| UD-Q3_K_XL | 16.6 GB | 18.3 GB | 24 GB cards |
| UD-IQ4_XS | 17.5 GB | 19.2 GB | 24 GB cards |
| UD-IQ4_NL | 17.8 GB | 19.5 GB | 24 GB cards |
| UD-Q4_K_L | 20.2 GB | 21.9 GB | 24 GB cards |
| Q4_K_S | 20.7 GB | 22.3 GB | 24 GB cards |
| MXFP4_MOE | 21.6 GB | 23.3 GB | 24 GB cards |
| Q4_K_M | 22.0 GB | 23.7 GB | 32 GB cards |
| UD-Q4_K_XL | 22.2 GB | 23.9 GB | 32 GB cards |
| Q5_K_S | 24.8 GB | 26.5 GB | 32 GB cards |
| Q5_K_M | 26.2 GB | 27.9 GB | 32 GB cards |
| UD-Q5_K_XL | 26.4 GB | 28.1 GB | 32 GB cards |
| UD-Q6_K_S | 28.5 GB | 30.2 GB | 32 GB cards |
| Q6_K | 28.9 GB | 30.5 GB | 32 GB cards |
| UD-Q6_K_XL | 32.1 GB | 33.7 GB | 48 GB cards |
| Q8_0 | 36.9 GB | 38.6 GB | 48 GB cards |
| UD-Q8_K_XL | 48.7 GB | 50.4 GB | 96 GB cards |
| BF16 | 69.4 GB | 71.0 GB | 96 GB cards |
Expected speed on popular hardware
| Hardware | Best quant that fits | Expected tok/s |
|---|---|---|
| NVIDIA RTX 3060 | Does not fit | |
| NVIDIA RTX 4060 Ti | Does not fit | |
| NVIDIA RTX 3090 | Q4_K_S | 185 estimate |
| NVIDIA RTX 4090 | Q4_K_S | 246 estimate |
| NVIDIA RTX 5090 | Q6_K | 232 estimate |
| AMD Radeon RX 7900 XTX | Q4_K_S | 206 estimate |
| Apple M4 Pro (20-core GPU) (64 GB) | Q8_0 | 36.6 estimate |
| Apple M3 Max (40-core GPU) (64 GB) | Q8_0 | 53.5 estimate |
| AMD Ryzen AI Max+ 395 (Radeon 8060S) | Q8_0 | 32.9 estimate |
Too big for your card?
Run Qwen3.5 35B-A3B on a rented 24-80 GB GPU by the hour.
Rent a bigger GPU on RunPodAffiliate link: if you sign up, RigBench may earn a referral credit. It never affects rankings or numbers.