Can RTX 4090 run Qwen3 30B-A3B Instruct?
Comfortable: GGUF Q4_K_M needs 19.7 GB at 4K, batch 1; headroom 4.3 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
| Framework | Quant | tok/s | VRAM used (GB) |
|---|---|---|---|
| llama.cpp | GGUF Q4_K_M | 95 | 19 |
Quant memory at 4K
| Quant | Total GB | Verdict | Weight sizing |
|---|---|---|---|
| GGUF Q4_K_M | 19.7 | Comfortable | Estimated weights |
| GGUF Q5_K_M | 23.0 | Tight | Estimated weights |
| AWQ INT4 | 16.4 | Comfortable | Estimated weights |
Context capacity · GGUF Q4_K_M
| Context | Total GB | Verdict |
|---|---|---|
| 4K | 19.7 | Comfortable |
| 16K | 21.0 | Comfortable |
| 32K | 22.6 | Tight |
Generation speed
Measured: 95 tok/s, GGUF Q4_K_M, llama.cpp
GGUF Q4_K_M: 95 tok/s · Measured