QwQ 32B at Q4_K_M needs about 22.4 GB, more than the 12 GB of the RTX 5070: part of the model runs from system RAM, which slows generation. At Q3_K_M it does not fit in VRAM.
Q4_K_M · 8K context
The model needs about 22.4 GB but only 11.6 GB of VRAM is usable: 10.7 GB must be loaded into system RAM. It runs, 52% on the GPU, but more slowly.
Estimated speed is at most 3.5 tokens/s: usable, but slow.
~22.4 GB (21.6 GB–23.2 GB) needed · estimate
10.7 GB must be offloaded to system RAM
EstimateData updated 24 days agoSource: Run LLM Locally engine, Hugging Face config.json + model card
What makes this estimate less reliable
| QwQ 32B Q8_08K~37 GB · 0.8–1.5 tok/s estimate | Not recommended |
|---|---|
| QwQ 32B Q6_K8K~29.3 GB · 1.1–2 tok/s estimate | Possible but slow |
| QwQ 32B Q5_K_M8K~25.7 GB · 1.4–2.5 tok/s estimate | Possible but slow |
| QwQ 32B Q4_K_M8K~22.4 GB · 1.9–3.5 tok/s estimate | Possible but slow |
| QwQ 32B IQ4_XS8K~20.3 GB · 2.5–4.5 tok/s estimate | Possible but slow |
| QwQ 32B Q3_K_M8K~18.5 GB · 3–5.5 tok/s estimate | CPU/GPU offload |
Prices and links follow the market you choose.
AMD
CompatibleRuns QwQ 32B Q4_K_M entirely in memory
Apple Silicon
CompatibleRuns QwQ 32B Q4_K_M entirely in memory
Unified memory PC
CompatibleRuns QwQ 32B Q4_K_M entirely in memory
No current offer imported for this market: the link opens a marketplace search.
