Qwen on NVIDIA Ampere

Can I run Qwen2.5-Coder 32B on an A100 80GB?

Yes. Qwen2.5-Coder 32B at Q4_K_M uses 21.5 GB of the 74.4 GB available on a A100 80GB, and runs at about 63.5 tokens per second. You can push the context to 128K.

Runs comfortably

21.5 GB of 74.4 GB · 29%
074 GB
Weights 18.6 GB
KV cache 2.0 GB
Runtime overhead 0.8 GB

Qwen2.5-Coder 32B at Q4_K_M leaves 52.9 GB spare on a A100 80GB. There is room to raise the context length or move up a quantisation level.

Generation63.5tok/s
Prompt processing2000tok/s
Max context128Ktokens
KV per 1K tokens0GB

Every quantisation of Qwen2.5-Coder 32B on a A100 80GB

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 Runs comfortably 50K 20.4
INT8 / W8A8 8.50 32.1 GB 34.9 GB Runs comfortably 128K 38.0
Q8_0 (GGUF) 8.50 32.1 GB 34.9 GB Runs comfortably 128K 38.0
FP8 (E4M3) 8.00 30.5 GB 33.3 GB Runs comfortably 128K 39.9
Q6_K 6.56 25.0 GB 27.9 GB Runs comfortably 128K 48.1
Q5_K_M 5.67 21.6 GB 24.5 GB Runs comfortably 128K 55.2
Q5_K_S 5.52 21.1 GB 23.9 GB Runs comfortably 128K 56.6
Q4_K_M 4.85 18.6 GB 21.5 GB Runs comfortably 128K 63.5
AWQ 4-bit 4.25 18.3 GB 21.2 GB Runs comfortably 128K 64.4
GPTQ 4-bit 4.25 18.3 GB 21.2 GB Runs comfortably 128K 64.4
MXFP4 4.25 18.3 GB 21.2 GB Runs comfortably 128K 64.4
Q4_K_S 4.58 17.6 GB 20.5 GB Runs comfortably 128K 66.7
Q4_0 4.55 17.5 GB 20.4 GB Runs comfortably 128K 67.1
IQ4_XS 4.25 16.4 GB 19.3 GB Runs comfortably 128K 71.2
Q3_K_M 3.91 15.2 GB 18.0 GB Runs comfortably 128K 76.5
IQ3_M 3.70 14.4 GB 17.3 GB Runs comfortably 128K 80.1
IQ3_XXS 3.06 12.1 GB 15.0 GB Runs comfortably 128K 93.8
Q2_K 2.63 10.6 GB 13.4 GB Runs comfortably 128K 106
IQ2_XXS 2.06 8.5 GB 11.3 GB Runs comfortably 128K 128
IQ1_M 1.75 7.4 GB 10.2 GB Runs comfortably 128K 144

Also worth checking

Qwen2.5-Coder 32B on every GPU
Full hardware tableEverything that runs on a A100 80GB
Full model tableWhy the KV cache is the part that bites
Guide