Qwen on NVIDIA Hopper

Can I run Qwen3 4B on an H100 SXM 80GB?

Yes. Qwen3 4B at Q4_K_M uses 4.2 GB of the 74.4 GB available on a H100 SXM 80GB, and runs at about 697 tokens per second. You can push the context to 128K.

Runs comfortably

4.2 GB of 74.4 GB · 6%
074 GB
Weights 2.3 GB
KV cache 1.1 GB
Runtime overhead 0.8 GB

Qwen3 4B at Q4_K_M leaves 70.2 GB spare on a H100 SXM 80GB. There is room to raise the context length or move up a quantisation level.

Generation697tok/s
Prompt processing51664tok/s
Max context128Ktokens
KV per 1K tokens0GB

Every quantisation of Qwen3 4B on a H100 SXM 80GB

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

QuantisationbpwWeights TotalVerdictMax ctxtok/s
FP16 / BF16 16.00 7.5 GB 9.4 GB Runs comfortably 128K 274
INT8 / W8A8 8.50 3.9 GB 5.8 GB Runs comfortably 128K 473
Q8_0 (GGUF) 8.50 3.9 GB 5.8 GB Runs comfortably 128K 473
FP8 (E4M3) 8.00 3.7 GB 5.7 GB Runs comfortably 128K 487
Q6_K 6.56 3.1 GB 5.0 GB Runs comfortably 128K 567
Q5_K_M 5.67 2.7 GB 4.6 GB Runs comfortably 128K 630
Q5_K_S 5.52 2.6 GB 4.5 GB Runs comfortably 128K 642
AWQ 4-bit 4.25 2.5 GB 4.4 GB Runs comfortably 128K 653
GPTQ 4-bit 4.25 2.5 GB 4.4 GB Runs comfortably 128K 653
MXFP4 4.25 2.5 GB 4.4 GB Runs comfortably 128K 653
Q4_K_M 4.85 2.3 GB 4.2 GB Runs comfortably 128K 697
Q4_K_S 4.58 2.2 GB 4.1 GB Runs comfortably 128K 721
Q4_0 4.55 2.2 GB 4.1 GB Runs comfortably 128K 724
IQ4_XS 4.25 2.0 GB 4.0 GB Runs comfortably 128K 753
Q3_K_M 3.91 1.9 GB 3.8 GB Runs comfortably 128K 790
IQ3_M 3.70 1.8 GB 3.7 GB Runs comfortably 128K 814
IQ3_XXS 3.06 1.5 GB 3.5 GB Runs comfortably 128K 898
Q2_K 2.63 1.4 GB 3.3 GB Runs comfortably 128K 965
IQ2_XXS 2.06 1.1 GB 3.0 GB Runs comfortably 128K 1072
IQ1_M 1.75 1.0 GB 2.9 GB Runs comfortably 128K 1140

Also worth checking

Qwen3 4B on every GPU
Full hardware tableEverything that runs on a H100 SXM 80GB
Full model tableWhy the KV cache is the part that bites
Guide