Model Anatomy See what's inside open-weight AI models

Can I run QwQ-32B on an RTX 6000 Ada?

Yes, with a GGUF version

The published weights are too big, but the Q8_0 file from Qwen/QwQ-32B-GGUF (34.8 GB) fits, with room for ≈ 31K tokens of context.

Full analysis of QwQ-32B Everything that runs on RTX 6000 Ada Get the Q8_0 file on Hugging Face
ollama run hf.co/Qwen/QwQ-32B-GGUF:Q8_0

Files and context on RTX 6000 Ada

Each file of QwQ-32B (and of GGUF versions the site has analyzed), and how much context fits next to it, with the context cache in 16-bit or 8-bit.

FileSizeOn RTX 6000 Ada Context (16-bit)Context (8-bit)
As published65.5 GBDoesn't fit——
qwq-32b-fp16Qwen/QwQ-32B-GGUF65.5 GBDoesn't fit——
Q8_0Qwen/QwQ-32B-GGUF34.8 GBTight31K62K
Q6_KQwen/QwQ-32B-GGUF26.9 GBFits61K122K
Q5_K_MQwen/QwQ-32B-GGUF23.3 GBFits74K128K
Q5_0Qwen/QwQ-32B-GGUF22.6 GBFits77K128K
Q4_K_MQwen/QwQ-32B-GGUF19.9 GBFits87K128K

QwQ-32B on other hardware

Popular models that run on RTX 6000 Ada

Assumes 90% of the card's memory is usable and at least 4K tokens of context for one conversation, with the context cache in 16-bit unless noted. Fits leaves at least 20% of that memory for context; Tight leaves less.