Can the RTX 5090 run Qwen3.6 35B A3B?
Yes. Qwen3.6 35B A3B fits the RTX 5090's 32 GB at Q4_K_M (about 22.8 GB needed), running at an estimated 391 tok/s. There's room for up to Q5_K_M if you want higher quality.
Every quantization
Qwen3.6 35B A3B (36B, 3B active) on 32 GB
| Quant | Weights | Needs | Fit | Est. speed |
|---|---|---|---|---|
| Q4_K_M | 21 GB | 22.8 GB | Fits | ~391 tok/s |
| Q5_K_M | 26 GB | 27.8 GB | Tight | ~316 tok/s |
| Q8_0 | 39 GB | 40.8 GB | Too big | - |
| FP16 | 72 GB | 73.8 GB | Too big | - |
Estimates. Needs = weights + 1.8 GB for KV cache and runtime at modest context. Speed is single-stream decode scaled from the RTX 5090's 1792 GB/s bandwidth (Estimated from memory bandwidth (llama.cpp decode)). Long contexts need more memory and run slower.
Run it
Copy, paste, go
llama-server -hf unsloth/Qwen3.6-35B-A3B-GGUF -ngl 99Ollama pulls a Q4 build by default. Pick a larger quant only if the table above says it fits. Full Qwen3.6 35B A3B details →
Compare
Same model, other hardware - and other models on this card
Qwen3.6 35B A3B on other hardware
- RTX 5090 D~391 tok/s
- RTX 6000D~343 tok/s
- RTX 4090~238 tok/s
- Radeon RX 7900 XTX~210 tok/s
- RTX 3090~209 tok/s
- MacBook Pro 16" M4 Max~119 tok/s
- Arc Pro B60~100 tok/s
- DGX Spark (GB10)~60 tok/s
Also runs on the RTX 5090
- Qwen3.5 35B A3B~391 tok/s
- dolphin-2.9.1-yi-1.5-34b~51 tok/s
- Gemma 4 31B~54 tok/s
- Qwen3 32B~54 tok/s
- GLM-4.7-Flash~396 tok/s
- Qwen3-30B-A3B-Instruct-2507~387 tok/s
- Qwen3.6 27B~64 tok/s
- Qwen3.8 27B~64 tok/s
Frequently asked
Quick answers
Can the NVIDIA RTX 5090 run Qwen3.6 35B A3B?
Yes. Qwen3.6 35B A3B fits the RTX 5090's 32 GB at Q4_K_M (about 22.8 GB needed), running at an estimated 391 tok/s. There's room for up to Q5_K_M if you want higher quality.
How much VRAM does Qwen3.6 35B A3B need?
About 21 GB of weights at Q4_K_M, 39 GB at Q8_0 and 72 GB at FP16. Add roughly 1.8 GB for the KV cache and runtime at modest context lengths.
How fast is Qwen3.6 35B A3B on the RTX 5090?
Roughly 391 tokens per second at Q4_K_M (single user, llama.cpp-style decode). This is an estimate from the RTX 5090's 1792 GB/s memory bandwidth; as a mixture-of-experts model only about 3B parameters are read per token, which is why it is faster than its size suggests.