Qwen on NVIDIA Ampere

Can I run Qwen2.5-Coder 32B on an RTX 3090?

Yes. Qwen2.5-Coder 32B at Q4_K_M uses 21.5 GB of the 21.8 GB available on a RTX 3090, and runs at about 29.5 tokens per second. You can push the context to 9K.

Fits, but tight

21.5 GB of 21.8 GB · 99%
022 GB
Weights 18.6 GB
KV cache 2.0 GB
Runtime overhead 0.8 GB

This fits with almost nothing to spare. A background application claiming VRAM will push it over. Drop to the next quantisation down, or quantise the KV cache to Q8_0 — that halves the cache for no meaningful quality loss.

Generation29.5tok/s
Prompt processing455tok/s
Max context9Ktokens
KV per 1K tokens0GB

Every quantisation of Qwen2.5-Coder 32B on a RTX 3090

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 — 0.89
INT8 / W8A8 8.50 32.1 GB 34.9 GB Won't fit — 2.62
Q8_0 (GGUF) 8.50 32.1 GB 34.9 GB Won't fit — 2.62
FP8 (E4M3) 8.00 30.5 GB 33.3 GB Won't fit — 2.93
Q6_K 6.56 25.0 GB 27.9 GB Won't fit — 5.09
Q5_K_M 5.67 21.6 GB 24.5 GB Won't fit — 9.53
Q5_K_S 5.52 21.1 GB 23.9 GB Won't fit — 10.6
Q4_K_M 4.85 18.6 GB 21.5 GB Fits, but tight 9K 29.5
AWQ 4-bit 4.25 18.3 GB 21.2 GB Fits, but tight 10K 29.9
GPTQ 4-bit 4.25 18.3 GB 21.2 GB Fits, but tight 10K 29.9
MXFP4 4.25 18.3 GB 21.2 GB Fits, but tight 10K 29.9
Q4_K_S 4.58 17.6 GB 20.5 GB Fits, but tight 13K 31.1
Q4_0 4.55 17.5 GB 20.4 GB Fits, but tight 14K 31.2
IQ4_XS 4.25 16.4 GB 19.3 GB Runs comfortably 18K 33.2
Q3_K_M 3.91 15.2 GB 18.0 GB Runs comfortably 23K 35.7
IQ3_M 3.70 14.4 GB 17.3 GB Runs comfortably 26K 37.4
IQ3_XXS 3.06 12.1 GB 15.0 GB Runs comfortably 35K 43.9
Q2_K 2.63 10.6 GB 13.4 GB Runs comfortably 41K 49.8
IQ2_XXS 2.06 8.5 GB 11.3 GB Runs comfortably 50K 60.4
IQ1_M 1.75 7.4 GB 10.2 GB Runs comfortably 54K 68.3

Also worth checking

Qwen2.5-Coder 32B on every GPU
Full hardware tableEverything that runs on a RTX 3090
Full model tableWhy the KV cache is the part that bites
Guide