Microsoft on NVIDIA Hopper

Can I run Phi-4 14B on an H100 SXM 80GB?

Yes. Phi-4 14B at Q4_K_M uses 10.8 GB of the 74.4 GB available on a H100 SXM 80GB, and runs at about 241 tokens per second. You can push the context to 16K.

Runs comfortably

10.8 GB of 74.4 GB · 14%
074 GB
Weights 8.4 GB
KV cache 1.6 GB
Runtime overhead 0.8 GB

Phi-4 14B at Q4_K_M leaves 63.6 GB spare on a H100 SXM 80GB. There is room to raise the context length or move up a quantisation level.

Generation241tok/s
Prompt processing14129tok/s
Max context16Ktokens
KV per 1K tokens0GB

Every quantisation of Phi-4 14B 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 27.4 GB 29.8 GB Runs comfortably 16K 81.7
INT8 / W8A8 8.50 14.3 GB 16.7 GB Runs comfortably 16K 150
Q8_0 (GGUF) 8.50 14.3 GB 16.7 GB Runs comfortably 16K 150
FP8 (E4M3) 8.00 13.7 GB 16.1 GB Runs comfortably 16K 156
Q6_K 6.56 11.2 GB 13.6 GB Runs comfortably 16K 187
Q5_K_M 5.67 9.7 GB 12.1 GB Runs comfortably 16K 212
Q5_K_S 5.52 9.4 GB 11.8 GB Runs comfortably 16K 217
AWQ 4-bit 4.25 8.7 GB 11.1 GB Runs comfortably 16K 234
GPTQ 4-bit 4.25 8.7 GB 11.1 GB Runs comfortably 16K 234
MXFP4 4.25 8.7 GB 11.1 GB Runs comfortably 16K 234
Q4_K_M 4.85 8.4 GB 10.8 GB Runs comfortably 16K 241
Q4_K_S 4.58 7.9 GB 10.3 GB Runs comfortably 16K 252
Q4_0 4.55 7.9 GB 10.3 GB Runs comfortably 16K 254
IQ4_XS 4.25 7.4 GB 9.8 GB Runs comfortably 16K 268
Q3_K_M 3.91 6.9 GB 9.3 GB Runs comfortably 16K 285
IQ3_M 3.70 6.5 GB 8.9 GB Runs comfortably 16K 297
IQ3_XXS 3.06 5.5 GB 7.9 GB Runs comfortably 16K 341
Q2_K 2.63 4.8 GB 7.2 GB Runs comfortably 16K 379
IQ2_XXS 2.06 3.9 GB 6.3 GB Runs comfortably 16K 444
IQ1_M 1.75 3.4 GB 5.8 GB Runs comfortably 16K 490

Also worth checking

Phi-4 14B 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