Models › Qwen3 Coder 30B-A3B Instruct
Qwen3 Coder 30B-A3B Instruct: VRAM needs and speed by GPU
We have 2 measured speed results for it.
30.53B parameters (mixture-of-experts, ~3.1B active) · license: apache-2.0 · 262,144-token context · GGUF published Jul 31, 2025 · GGUF files on Hugging Face
Measured speeds
| Hardware | Generate tok/s | Prompt tok/s | Quant | Backend | Test | Source |
|---|---|---|---|---|---|---|
| Apple M2 Ultra | 79.0 | 2,453 | Q8_0 | Metal FA on | tg32 | llama.cpp benches details |
| NVIDIA DGX Spark (GB10) | 61.1 | 2,987 | Q8_0 | CUDA FA on | tg32 | llama.cpp benches details |
Memory needed per quantization
| Quant | File size | Needs (estimate) | Smallest card it fits |
|---|---|---|---|
| Q8_0 | 32.5 GB | 34.0 GB | 48 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 | Does not fit | |
| NVIDIA RTX 4090 | Does not fit | |
| NVIDIA RTX 5090 | Does not fit | |
| AMD Radeon RX 7900 XTX | Does not fit | |
| Apple M4 Pro (20-core GPU) (64 GB) | Q8_0 | 35.6 estimate |
| Apple M3 Max (40-core GPU) (64 GB) | Q8_0 | 52.0 estimate |
| AMD Ryzen AI Max+ 395 (Radeon 8060S) | Q8_0 | 32.0 estimate |
Too big for your card?
Run Qwen3 Coder 30B-A3B Instruct 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.