Qwen on NVIDIA Ampere

Can I run Qwen2.5 72B on an A100 80GB?

Yes. Qwen2.5 72B at Q4_K_M uses 44.6 GB of the 74.4 GB available on a A100 80GB, and runs at about 29.6 tokens per second. You can push the context to 103K.

Runs comfortably

44.6 GB of 74.4 GB · 60%
074 GB
Weights 41.2 GB
KV cache 2.5 GB
Runtime overhead 0.9 GB

Qwen2.5 72B at Q4_K_M leaves 29.8 GB spare on a A100 80GB. There is room to raise the context length or move up a quantisation level.

Generation29.6tok/s
Prompt processing901tok/s
Max context103Ktokens
KV per 1K tokens0GB

Every quantisation of Qwen2.5 72B on a A100 80GB

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

QuantisationbpwWeights TotalVerdictMax ctxtok/s
FP16 / BF16 16.00 135.4 GB 138.8 GB Won't fit — 0.58
INT8 / W8A8 8.50 71.4 GB 74.8 GB Won't fit 7K 12.6
Q8_0 (GGUF) 8.50 71.4 GB 74.8 GB Won't fit 7K 12.6
FP8 (E4M3) 8.00 67.7 GB 71.1 GB Fits, but tight 19K 18.3
Q6_K 6.56 55.5 GB 58.9 GB Runs comfortably 58K 22.2
Q5_K_M 5.67 48.0 GB 51.4 GB Runs comfortably 82K 25.6
Q5_K_S 5.52 46.7 GB 50.1 GB Runs comfortably 86K 26.3
Q4_K_M 4.85 41.2 GB 44.6 GB Runs comfortably 103K 29.6
AWQ 4-bit 4.25 39.4 GB 42.8 GB Runs comfortably 109K 30.9
GPTQ 4-bit 4.25 39.4 GB 42.8 GB Runs comfortably 109K 30.9
MXFP4 4.25 39.4 GB 42.8 GB Runs comfortably 109K 30.9
Q4_K_S 4.58 39.0 GB 42.4 GB Runs comfortably 110K 31.2
Q4_0 4.55 38.8 GB 42.2 GB Runs comfortably 111K 31.4
IQ4_XS 4.25 36.3 GB 39.7 GB Runs comfortably 119K 33.4
Q3_K_M 3.91 33.6 GB 36.9 GB Runs comfortably 128K 36.0
IQ3_M 3.70 31.8 GB 35.2 GB Runs comfortably 128K 37.9
IQ3_XXS 3.06 26.6 GB 30.0 GB Runs comfortably 128K 44.8
Q2_K 2.63 23.1 GB 26.5 GB Runs comfortably 128K 51.1
IQ2_XXS 2.06 18.4 GB 21.8 GB Runs comfortably 128K 62.9
IQ1_M 1.75 15.9 GB 19.3 GB Runs comfortably 128K 71.9

Also worth checking

Qwen2.5 72B on every GPU
Full hardware tableEverything that runs on a A100 80GB
Full model tableWhy the KV cache is the part that bites
Guide