ToolSite

What LLMs can eight H100 SXM GPUs run?

Eight H100 SXM GPUs give a model 640 GB (8 × 80 GB) of memory and 26,800 GB/s of bandwidth in total. Of the 40 open models tracked here, 37 fit at Q4_K_M with an 8,192-token context. The speeds assume tensor parallelism, as in vLLM or SGLang; llama.cpp splits layers across cards by default and then writes at about the speed of one card.

Models that fit on 8× H100 SXM

The most precise weights that still fit with 8K tokens of context and one request, the memory that takes, the longest context at that precision and the writing speed. Bigger models come first.

Model Best precision Memory Longest context Tokens/s
MiMo V2.6 Pro 1.02T, 42B active GGUF Q4_K_M 636 GB 21K 149–310
DeepSeek V4.1 Flash 763.2B, 16B active GGUF Q5_K_M 555 GB 952K 200–455
GLM-5.3 753.3B, 40B active GGUF Q6_K 634 GB 32K 130–263
GLM-5.2 753.3B, 40B active GGUF Q6_K 634 GB 32K 130–263
DeepSeek V3.2 685.4B, 37B active GGUF Q6_K 577 GB 160K (full) 136–277
DeepSeek V3 / R1 684.5B, 37B active GGUF Q6_K 576 GB 160K (full) 136–277
MiniMax M3 427.0B, 23B active GGUF Q8_0 466 GB 1M (full) 150–313
GLM-5.3 Flash 321.3B, 18B active GGUF Q8_0 350 GB 1M (full) 170–366
MiMo V2.6 Flash 310.8B, 15B active GGUF Q8_0 339 GB 1M (full) 182–399
DeepSeek V4 Flash 0731 304.2B, 13B active FP16 / BF16 624 GB 138K 147–304
DeepSeek V4 Flash 290.9B, 13B active FP16 / BF16 597 GB 431K 147–304
MiniMax M2.7 228.7B, 10B active FP16 / BF16 471 GB 200K (full) 160–339
Qwen3.8 Flash Next 180.0B, 6B active FP16 / BF16 370 GB 256K (full) 199–452
Mistral Medium 3.5 128B 127.7B FP16 / BF16 265 GB 256K (full) 48–71
Qwen3.5 122B-A10B 125.1B, 10B active FP16 / BF16 257 GB 256K (full) 166–356
Nemotron 3 Super 120B-A12B 123.6B, 12B active FP16 / BF16 254 GB 256K (full) 154–323
gpt-oss-120b 116.8B, 5.1B active FP16 / BF16 240 GB 128K (full) 208–480
Qwen3-Coder-Next 79.7B, 3B active FP16 / BF16 164 GB 256K (full) 234–567
Llama 3.1 70B 70.6B FP16 / BF16 148 GB 128K (full) 75–118
Qwen3.6 35B-A3B 36.0B, 3B active FP16 / BF16 74.3 GB 256K (full) 234–568
Nemotron 3 Nano 30B-A3B 31.6B, 3.5B active FP16 / BF16 65.3 GB 256K (full) 228–548
Gemma 4 31B 31.3B FP16 / BF16 65.8 GB 256K (full) 128–224
GLM-4.7 Flash 31.2B, 3B active FP16 / BF16 64.9 GB 198K (full) 232–562
Xing 4.0 29B-A4B 31.2B, 4B active FP16 / BF16 64.8 GB 256K (full) 220–519
Qwen3 30B-A3B 30.5B, 3.3B active FP16 / BF16 63.9 GB 40K (full) 226–540
Muse Glimmer 30B 29.8B FP16 / BF16 61.7 GB 128K (full) 132–234
Qwen3.6 27B 27.8B FP16 / BF16 58.0 GB 256K (full) 137–244
Qwen3.8 27B 27.8B FP16 / BF16 58.0 GB 256K (full) 137–244
Gemma 4 26B-A4B 25.8B, 4B active FP16 / BF16 53.7 GB 256K (full) 221–521
gpt-oss-20b 20.9B, 3.6B active FP16 / BF16 43.6 GB 128K (full) 226–540
Gemma 4 12B 12.0B FP16 / BF16 25.4 GB 256K (full) 194–398
Qwen3.5 9B 9.7B FP16 / BF16 20.6 GB 256K (full) 207–440
Qwen3 8B 8.2B FP16 / BF16 18.5 GB 40K (full) 213–460
Llama 3.1 8B 8.0B FP16 / BF16 18.1 GB 128K (full) 214–465
Gemma 4 E4B 8.0B FP16 / BF16 17.0 GB 128K (full) 218–475
Nemotron 3 Nano 4B 4.0B FP16 / BF16 8.78 GB 256K (full) 247–588
MiniCPM5 2B 2.5B FP16 / BF16 6.02 GB 128K (full) 259–638

Too big for 8× H100 SXM

How many H100 SXM cards these need at Q4_K_M in one tensor-parallel group; some need more than 8.

How these numbers are worked out

Memory is the weights at each precision plus the KV cache for 8,192 tokens in FP16 and runtime overhead (0.5 GB plus 10%), computed from each model's files on Hugging Face with the LLM VRAM Calculator. Speed is estimated from memory bandwidth and the parameters read per token with the LLM Speed Calculator. Both tools take any other model from Hugging Face.

Other GPUs and Macs

Model numbers read from Hugging Face on .