Llama on NVIDIA Hopper

Can I run Llama 4 Scout 109B-A17B on an H100 SXM 80GB?

Yes. Llama 4 Scout 109B-A17B at Q4_K_M uses 64.0 GB of the 74.4 GB available on a H100 SXM 80GB, and runs at about 177 tokens per second. You can push the context to 63K.

Runs comfortably

64.0 GB of 74.4 GB · 86%
074 GB
Weights 61.7 GB
KV cache 1.5 GB
Runtime overhead 0.8 GB

Llama 4 Scout 109B-A17B at Q4_K_M leaves 10.4 GB spare on a H100 SXM 80GB. There is room to raise the context length or move up a quantisation level.

Generation177tok/s
Prompt processing12217tok/s
Max context63Ktokens
KV per 1K tokens0GB

Every quantisation of Llama 4 Scout 109B-A17B on a H100 SXM 80GB

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

QuantisationbpwWeights TotalVerdictMax ctxtok/s
FP16 / BF16 16.00 203.0 GB 205.4 GB Won't fit — 2.11
INT8 / W8A8 8.50 107.4 GB 109.7 GB Won't fit — 7.31
Q8_0 (GGUF) 8.50 107.4 GB 109.7 GB Won't fit — 7.31
FP8 (E4M3) 8.00 101.5 GB 103.9 GB Won't fit — 8.73
Q6_K 6.56 83.2 GB 85.6 GB Won't fit — 19.6
Q5_K_M 5.67 71.9 GB 74.3 GB Fits, but tight 9K 155
Q5_K_S 5.52 70.0 GB 72.4 GB Fits, but tight 19K 159
Q4_K_M 4.85 61.7 GB 64.0 GB Runs comfortably 63K 177
Q4_K_S 4.58 58.3 GB 60.7 GB Runs comfortably 81K 185
Q4_0 4.55 58.0 GB 60.3 GB Runs comfortably 83K 186
AWQ 4-bit 4.25 56.8 GB 59.1 GB Runs comfortably 90K 190
GPTQ 4-bit 4.25 56.8 GB 59.1 GB Runs comfortably 90K 190
MXFP4 4.25 56.8 GB 59.1 GB Runs comfortably 90K 190
IQ4_XS 4.25 54.2 GB 56.6 GB Runs comfortably 103K 197
Q3_K_M 3.91 50.0 GB 52.3 GB Runs comfortably 126K 211
IQ3_M 3.70 47.4 GB 49.7 GB Runs comfortably 140K 220
IQ3_XXS 3.06 39.4 GB 41.8 GB Runs comfortably 182K 255
Q2_K 2.63 34.1 GB 36.4 GB Runs comfortably 211K 285
IQ2_XXS 2.06 27.0 GB 29.3 GB Runs comfortably 248K 337
IQ1_M 1.75 23.1 GB 25.4 GB Runs comfortably 269K 375

Also worth checking

Llama 4 Scout 109B-A17B 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