Skip to content

Can RTX 4090 run Qwen3-Coder 30B-A3B Instruct?

Comfortable: GGUF Q4_K_M needs 19.8 GB at 4K, batch 1; headroom 4.2 GB.

Text input, FP16 KV cache, one card. Unified-memory capacity uses the GPU guide’s usable limit; negative headroom may require host offload. Speed source settings may differ from this 4K estimate.

Measured runs

FrameworkQuanttok/sVRAM used (GB)
llama.cppGGUF Q4_K_M9219.2

Quant memory at 4K

QuantTotal GBVerdictWeight sizing
GGUF Q4_K_M19.8ComfortableVerified file
GGUF Q5_K_M23.1TightVerified file
AWQ INT416.4ComfortableEstimated weights

Context capacity · GGUF Q4_K_M

ContextTotal GBVerdict
4K19.8Comfortable
16K21.0Comfortable
32K22.7Tight
128K32.6Does not fit

Generation speed

Measured: 92 tok/s, GGUF Q4_K_M, llama.cpp

GGUF Q4_K_M: 92 tok/s · Measured

Other model / GPU pairs