What GPU do you need to run Qwen3 Coder 30B-A3B?
30.5B params 256K max context Tool use
The smallest GPU memory that runs Qwen3 Coder 30B-A3B well is 24 GB (Q4_K_M, e.g. RTX Spark). 45 of 132 GPUs run it well and 2 more are a tight fit. It's a mixture-of-experts model, so all weights must fit in memory but speed depends only on the active experts.
MMLU 81
VRAM needed per quantization
| Quantization | Weights | 16K context | 32K context | Full 256K |
|---|---|---|---|---|
| Q4_K_M | 18.6 GB | 20.1 GB | 21.5 GB | 42.2 GB |
| Q8_0 | 32.5 GB | 34.0 GB | 35.4 GB | 56.1 GB |
Weights plus fp16 KV cache. Leave ~1 GB headroom for the inference engine.
Cheapest GPUs that can run Qwen3 Coder 30B-A3B
Lowest launch MSRP among GPUs that run its smallest quantization (Q4_K_M) well, with at least 16K context.
As an Amazon Associate, I earn from qualifying purchases.
GPUs that can run Qwen3 Coder 30B-A3B
Best quantization that fits on each GPU, with the context and speed it gets. For chips with several memory sizes, the smallest that works is shown.
NVIDIA
- RTX 4000 Ada 20 GB Q4_K_M 15K context · ~127 tok/s tight fit
- RTX 3090 24 GB Q4_K_M 58K context · ~331 tok/s
- RTX 3090 Ti 24 GB Q4_K_M 58K context · ~356 tok/s
- RTX 4090 24 GB Q4_K_M 58K context · ~356 tok/s
- RTX 4500 Ada 24 GB Q4_K_M 58K context · ~153 tok/s
- RTX A5000 24 GB Q4_K_M 58K context · ~271 tok/s
- RTX PRO 4000 Blackwell 24 GB Q4_K_M 58K context · ~237 tok/s
- RTX Spark 24 GB Q4_K_M 58K context · ~106 tok/s
- Tesla P40 24 GB Q4_K_M 58K context · ~122 tok/s
- RTX 5090 32 GB Q4_K_M 145K context · ~633 tok/s
- RTX 5000 Ada 32 GB Q4_K_M 145K context · ~204 tok/s
- RTX PRO 4500 Blackwell 32 GB Q4_K_M 145K context · ~317 tok/s
- Tesla V100 32GB 32 GB Q4_K_M 145K context · ~318 tok/s
- L40S 48 GB Q8_0 168K context · ~206 tok/s
- RTX 6000 Ada 48 GB Q8_0 168K context · ~229 tok/s
- RTX A6000 48 GB Q8_0 168K context · ~183 tok/s
- RTX PRO 5000 Blackwell 48 GB Q8_0 168K context · ~320 tok/s
- RTX PRO 5000 Blackwell 72GB 72 GB Q8_0 256K context · ~320 tok/s
- RTX PRO 5500 Blackwell 84 GB Q8_0 256K context · ~333 tok/s
- RTX PRO 6000 Blackwell 96 GB Q8_0 256K context · ~427 tok/s
- DGX Spark (GB 10) 128 GB Q8_0 256K context · ~65 tok/s
AMD
- Radeon RX 7900 XT 20 GB Q4_K_M 15K context · ~283 tok/s tight fit
- Radeon RX 7900 XTX 24 GB Q4_K_M 58K context · ~339 tok/s
- Radeon AI PRO R9700 32 GB Q4_K_M 145K context · ~226 tok/s
- Ryzen AI Max (Strix Halo) 32 GB Q4_K_M 145K context · ~90 tok/s
- Ryzen AI Max 400 (Gorgon Halo) 64 GB Q8_0 256K context · ~65 tok/s
Intel
- Arc Pro B65 32 GB Q4_K_M 145K context · ~215 tok/s
- Arc Pro B70 32 GB Q4_K_M 145K context · ~215 tok/s
Apple
- M2 24 GB Q4_K_M 58K context · ~35 tok/s
- M3 24 GB Q4_K_M 58K context · ~35 tok/s
- M4 24 GB Q4_K_M 58K context · ~42 tok/s
- M4 Pro 24 GB Q4_K_M 58K context · ~96 tok/s
- M5 24 GB Q4_K_M 58K context · ~54 tok/s
- M5 Pro 24 GB Q4_K_M 58K context · ~108 tok/s
- M6 24 GB Q4_K_M 58K context · ~60 tok/s
- M1 Max 32 GB Q4_K_M 145K context · ~141 tok/s
- M1 Pro 32 GB Q4_K_M 145K context · ~71 tok/s
- M2 Max 32 GB Q4_K_M 145K context · ~141 tok/s
- M2 Pro 32 GB Q4_K_M 145K context · ~71 tok/s
- M3 Max 36 GB Q8_0 38K context · ~71 tok/s
- M3 Pro 36 GB Q8_0 38K context · ~36 tok/s
- M4 Max 36 GB Q8_0 38K context · ~98 tok/s
- M5 Max 36 GB Q8_0 38K context · ~110 tok/s
- M1 Ultra 64 GB Q8_0 256K context · ~190 tok/s
- M2 Ultra 64 GB Q8_0 256K context · ~190 tok/s
- M3 Ultra 96 GB Q8_0 256K context · ~195 tok/s
- M5 Ultra 96 GB Q8_0 256K context · ~286 tok/s