Can I run Qwen3-Coder-Next on an RTX 4070 12GB?
With --n-cpu-moe 39: the Q4_K_M GGUF keeps 11.7 GB on the RTX 4070 and 35.3 GB in system RAM, with the experts of 39 of its 48 layers moved there. Whole, Qwen3-Coder-Next needs about 50.7 GB at Q4_K_M with 32K tokens of context, 38.7 GB more than the RTX 4070 12GB holds. The smallest setup here that holds Qwen3-Coder-Next at Q4_K_M with 32K is 2× RTX 5090 (64 GB).
With offload the Q4_K_M GGUF with --n-cpu-moe 39 and 32K tokens of context
- On the card
- 11.7 GB
- RTX 4070
- 12 GB, 504 GB/s
- In RAM
- 35.3 GB
- Tokens/s
- 22–37 tokens/s
With the Q4_K_M GGUF, --n-cpu-moe 39 and 32K tokens of context it writes about 22–37 tokens/s for one request on an RTX 4070 12GB and dual-channel DDR5-5600.
Qwen3-Coder-Next on the RTX 4070 as the context fills
One request, FP16 KV cache, 0.5 GB plus 10% overhead; a minus sign is memory missing, and tight is less than 0.5 GB free.
| Context | Q4_K_M | Free | Tokens/s | Q8_0 | Free | Tokens/s |
|---|---|---|---|---|---|---|
| 4K | 50.0 GB | −38.0 GB | — | 87.3 GB | −75.3 GB | — |
| 8K | 50.1 GB | −38.1 GB | — | 87.4 GB | −75.4 GB | — |
| 16K | 50.3 GB | −38.3 GB | — | 87.6 GB | −75.6 GB | — |
| 32K | 50.7 GB | −38.7 GB | — | 88.0 GB | −76.0 GB | — |
| 64K | 51.5 GB | −39.5 GB | — | 88.9 GB | −76.9 GB | — |
| 128K | 53.2 GB | −41.2 GB | — | 90.5 GB | −78.5 GB | — |
| 256K (full) | 56.5 GB | −44.5 GB | — | 93.8 GB | −81.8 GB | — |
--n-cpu-moe for Qwen3-Coder-Next on the RTX 4070
With --n-cpu-moe 39 the Q4_K_M GGUF keeps 11.7 GB on the RTX 4070 and 35.3 GB in RAM, about 22–37 tokens/s with DDR5-5600. Measured Q4_K_M file, 1 GB of buffers; tokens/s by system RAM speed.
Qwen3-Coder-Next on more than one RTX 4070
| Cards | Best at 32K | Tokens/s | Q4_K_M longest | Q4_K_M tokens/s |
|---|---|---|---|---|
| 2× (24 GB) | Nothing fits | — | Does not fit | — |
| 4× (48 GB) | Q3_K_M | 138–281 | Does not fit | — |
Tensor parallel with the combined bandwidth, as in vLLM; llama.cpp splits layers by default and runs at about one card's speed.
Other options
- The next smaller setting, Q3_K_M, takes 41.2 GB at 32K, 29.2 GB over the RTX 4070; it does not fit with 0.5 GB to spare even with 1K tokens.
- Smallest setup for Q4_K_M at 32K: 2× RTX 5090 (64 GB).
Run Qwen3-Coder-Next on the RTX 4070 with llama-server
llama-server -hf unsloth/Qwen3-Coder-Next-GGUF:Q4_K_M -c 32768 -ngl 99 --n-cpu-moe 39Qwen3-Coder-Next-Q4_K_M.gguf, 48.5 GB, from unsloth/
Questions
Can I run Qwen3-Coder-Next on an RTX 4070 12GB?
With --n-cpu-moe 39: the Q4_K_M GGUF keeps 11.7 GB on the RTX 4070 and 35.3 GB in system RAM, with the experts of 39 of its 48 layers moved there. Whole, Qwen3-Coder-Next needs about 50.7 GB at Q4_K_M with 32K tokens of context, 38.7 GB more than the RTX 4070 12GB holds. The smallest setup here that holds Qwen3-Coder-Next at Q4_K_M with 32K is 2× RTX 5090 (64 GB).
How fast is Qwen3-Coder-Next on an RTX 4070 12GB?
With the Q4_K_M GGUF, --n-cpu-moe 39 and 32K tokens of context it writes about 22–37 tokens/s for one request on an RTX 4070 12GB and dual-channel DDR5-5600.
Does Qwen3-Coder-Next need --n-cpu-moe on an RTX 4070 12GB?
With --n-cpu-moe 39 the Q4_K_M GGUF keeps 11.7 GB on the RTX 4070 and 35.3 GB in RAM, about 22–37 tokens/s with DDR5-5600.
What does a second RTX 4070 12GB change for Qwen3-Coder-Next?
Even two RTX 4070 12GB cards (24 GB) do not hold Qwen3-Coder-Next at 32K.
Try other settings in the VRAM calculator, the speed calculator or the MoE offload planner. See also Qwen3-Coder-Next VRAM requirements, what LLMs an RTX 4070 12GB can run and every pair, or detect your own GPU. Model data checked .