Qwen on NVIDIA Blackwell

Can I run Qwen2.5-Coder 32B on an RTX 5090?

Yes. Qwen2.5-Coder 32B at Q4_K_M uses 21.5 GB of the 29.3 GB available on a RTX 5090, and runs at about 59.0 tokens per second. You can push the context to 39K.

Runs comfortably

21.5 GB of 29.3 GB · 73%
029 GB
Weights 18.6 GB
KV cache 2.0 GB
Runtime overhead 0.8 GB

Qwen2.5-Coder 32B at Q4_K_M leaves 7.8 GB spare on a RTX 5090. There is room to raise the context length or move up a quantisation level.

Generation59.0tok/s
Prompt processing1340tok/s
Max context39Ktokens
KV per 1K tokens0GB

Every quantisation of Qwen2.5-Coder 32B on a RTX 5090

Highlighted row is the highest quality that still fits at 8K context.

QuantisationbpwWeights TotalVerdictMax ctxtok/s
FP16 / BF16 16.00 61.0 GB 63.9 GB Won't fit — 1.15
INT8 / W8A8 8.50 32.1 GB 34.9 GB Won't fit — 5.88
Q8_0 (GGUF) 8.50 32.1 GB 34.9 GB Won't fit — 5.88
FP8 (E4M3) 8.00 30.5 GB 33.3 GB Won't fit — 7.80
Q6_K 6.56 25.0 GB 27.9 GB Fits, but tight 14K 44.7
Q5_K_M 5.67 21.6 GB 24.5 GB Runs comfortably 27K 51.3
Q5_K_S 5.52 21.1 GB 23.9 GB Runs comfortably 30K 52.6
Q4_K_M 4.85 18.6 GB 21.5 GB Runs comfortably 39K 59.0
AWQ 4-bit 4.25 18.3 GB 21.2 GB Runs comfortably 40K 59.8
GPTQ 4-bit 4.25 18.3 GB 21.2 GB Runs comfortably 40K 59.8
MXFP4 4.25 18.3 GB 21.2 GB Runs comfortably 40K 59.8
Q4_K_S 4.58 17.6 GB 20.5 GB Runs comfortably 43K 62.0
Q4_0 4.55 17.5 GB 20.4 GB Runs comfortably 44K 62.4
IQ4_XS 4.25 16.4 GB 19.3 GB Runs comfortably 48K 66.2
Q3_K_M 3.91 15.2 GB 18.0 GB Runs comfortably 53K 71.1
IQ3_M 3.70 14.4 GB 17.3 GB Runs comfortably 56K 74.5
IQ3_XXS 3.06 12.1 GB 15.0 GB Runs comfortably 65K 87.3
Q2_K 2.63 10.6 GB 13.4 GB Runs comfortably 72K 98.7
IQ2_XXS 2.06 8.5 GB 11.3 GB Runs comfortably 80K 119
IQ1_M 1.75 7.4 GB 10.2 GB Runs comfortably 84K 134

Also worth checking

Qwen2.5-Coder 32B on every GPU
Full hardware tableEverything that runs on a RTX 5090
Full model tableWhy the KV cache is the part that bites
Guide