NVIDIA H100 (80GB)
The NVIDIA H100 (80GB) features 80 GB of dedicated VRAM and 3350 GB/s memory bandwidth, supporting CUDA and TensorRT-LLM runtimes for full-precision and quantized inference.
Memory
80 GB
Usable Memory Ceiling
80 GB
Bandwidth
3350 GB/s
Max Inference Class
70B+ Models (Q4)
TDP
700W
VRAM Scaling by Context Length
Estimated memory requirements as context length increases. Curves crossing above the reference line indicate Out of Memory (OOM).
Precision:
Best Models to Run on NVIDIA H100 (80GB)
Inference throughput, memory compatibility, and generation speeds for open-weights LLMs on NVIDIA H100 (80GB).
Sort by:
Context:
KV Cache:
GPUs:
| Rank | Model | Parameters | Precision | TPS (Approx.) | TTFT | Can Run |
|---|---|---|---|---|---|---|
#33 | Qwen 3.8 27B | 27B | INT4 Q8 FP16 | 138 tok/s 85 tok/s 46 tok/s | ~347 ms ~351 ms ~362 ms | |
#48 | Agnes 3.0 Flash | 33B | INT4 Q8 FP16 | 118 tok/s 71 tok/s 36 tok/s | ~421 ms ~427 ms ~476 ms | |
#61 | Sarvam-105B | 106B (10.3B active) | INT4 | 266 tok/s | ~178 ms | |
#72 | K2 Horizon MoVA 36B A4B | 36B (4B active) | INT4 Q8 FP16 | 381 tok/s 305 tok/s 199 tok/s | ~70 ms ~71 ms ~79 ms | |
#75 | Ling 3.0 Flash VL | 124B (5.5B active) | INT4 | 339 tok/s | ~138 ms | |
#83 | Qwen3.5-27B | 27B | INT4 Q8 FP16 | 138 tok/s 85 tok/s 46 tok/s | ~347 ms ~351 ms ~362 ms | |
#84 | Qwen3.5-122B-A10B | 122B (10B active) | INT4 | 211 tok/s | ~198 ms | |
#92 | Sarvam-30B | 32B (2.4B active) | INT4 Q8 FP16 | 438 tok/s 375 tok/s 267 tok/s | ~50 ms ~50 ms ~55 ms | |
#94 | Qwen3.6 35B A3B | 35B (3B active) | INT4 Q8 FP16 | 415 tok/s 345 tok/s 236 tok/s | ~58 ms ~58 ms ~64 ms | |
#96 | Muse Glimmer 30B | 30B | INT4 Q8 FP16 | 127 tok/s 77 tok/s 41 tok/s | ~384 ms ~389 ms ~401 ms | |
#100 | Phi-4 Reasoning Plus | 14B | INT4 Q8 FP16 | 173 tok/s 124 tok/s 75 tok/s | ~185 ms ~187 ms ~193 ms | |
#101 | MiMo V2 Flash | 15B (309B active) | INT4 Q8 FP16 | 15 tok/s 8 tok/s 4 tok/s | ~3.8s ~3.8s ~3.9s | |
#106 | GLM-4.5-Air | 106B (12B active) | INT4 | 88 tok/s | ~223 ms | |
#108 | Gemma 4 26B A4B | 25.2B (3.8B active) | INT4 Q8 FP16 | 396 tok/s 317 tok/s 219 tok/s | ~65 ms ~65 ms ~67 ms | |
#109 | Gemma 4 12B | 11.95B | INT4 Q8 FP16 | 192 tok/s 140 tok/s 86 tok/s | ~159 ms ~161 ms ~165 ms | |
#113 | Qwen3.5-9B | 9B | INT4 Q8 FP16 | 290 tok/s 201 tok/s 120 tok/s | ~120 ms ~122 ms ~125 ms | |
#114 | Qwen3.5-35B-A3B | 35B (3B active) | INT4 Q8 FP16 | 415 tok/s 345 tok/s 236 tok/s | ~58 ms ~58 ms ~64 ms | |
#114 | Ling 3.0 Flash | 124B (5.1B active) | INT4 | 350 tok/s | ~133 ms | |
#118 | K2 Horizon 7B | 7B | INT4 Q8 FP16 | 330 tok/s 237 tok/s 147 tok/s | ~95 ms ~96 ms ~99 ms | |
#126 | Qwen3 Next 80B A3B | 80B (79B active) | INT4 | 44 tok/s | ~1.0s | |
#128 | NVIDIA Nemotron 3 Nano 30B-A3B | 3.5B (30B active) | INT4 Q8 FP16 | 122 tok/s 75 tok/s 41 tok/s | ~373 ms ~378 ms ~389 ms | |
#129 | Gemma 4 31B | 30.7B | INT4 Q8 FP16 | 125 tok/s 76 tok/s 41 tok/s | ~393 ms ~398 ms ~410 ms | |
#130 | GPT-OSS 120B | 117B (5.1B active) | INT4 | 89 tok/s | ~150 ms | |
#131 | Nemotron 3.5 Lightning | 30B (3B active) | INT4 Q8 FP16 | 228 tok/s 212 tok/s 163 tok/s | ~59 ms ~59 ms ~65 ms | |
#132 | Qwen3.5-4B | 4B | INT4 Q8 FP16 | 418 tok/s 326 tok/s 219 tok/s | ~57 ms ~58 ms ~59 ms | |
#133 | Qwen3-30B-A3B | 30B (3B active) | INT4 Q8 FP16 | 418 tok/s 347 tok/s 250 tok/s | ~57 ms ~57 ms ~58 ms | |
#136 | GPT-OSS 20B | 21B (3.6B active) | INT4 Q8 FP16 | 247 tok/s 221 tok/s 170 tok/s | ~62 ms ~63 ms ~64 ms | |
#146 | Granite 4.2 8B | 8B | INT4 Q8 FP16 | 309 tok/s 218 tok/s 132 tok/s | ~107 ms ~109 ms ~112 ms | |
#150 | Gemma 4 E4B | 8B | INT4 Q8 FP16 | 309 tok/s 218 tok/s 132 tok/s | ~107 ms ~109 ms ~112 ms | |
#151 | Nemotron 3.5 Lightning 30B A3B | 30B (3B active) | INT4 Q8 FP16 | 418 tok/s 347 tok/s 250 tok/s | ~57 ms ~57 ms ~58 ms | |
#153 | Qwen3-32B | 32B | INT4 Q8 FP16 | 121 tok/s 73 tok/s 37 tok/s | ~409 ms ~415 ms ~462 ms | |
#154 | Qwen2.5-72B | 72B | INT4 Q8 | 60 tok/s 33 tok/s | ~910 ms ~1.0s | |
#155 | Mistral Large 3 | 41B (675B active) | INT4 Q8 | 7 tok/s 4 tok/s | ~8.2s ~8.3s | |
#156 | GLM-4 | 32B | INT4 Q8 FP16 | 97 tok/s 64 tok/s 34 tok/s | ~411 ms ~416 ms ~464 ms | |
#157 | DeepSeek-R1 70B | 70B | INT4 Q8 | 63 tok/s 34 tok/s | ~884 ms ~973 ms | |
#158 | GLM-4V | 9B | INT4 Q8 FP16 | 230 tok/s 172 tok/s 109 tok/s | ~121 ms ~123 ms ~126 ms | |
#159 | Gemma 3 27B | 27B | INT4 Q8 FP16 | 138 tok/s 85 tok/s 46 tok/s | ~347 ms ~351 ms ~362 ms | |
#160 | Phi-4 | 14B | INT4 Q8 FP16 | 221 tok/s 145 tok/s 83 tok/s | ~184 ms ~186 ms ~192 ms | |
#161 | Llama 3.3 70B | 70B | INT4 Q8 | 62 tok/s 34 tok/s | ~885 ms ~973 ms | |
#162 | DeepSeek-R1 32B | 32B | INT4 Q8 FP16 | 124 tok/s 74 tok/s 37 tok/s | ~409 ms ~414 ms ~462 ms | |
#163 | Hunyuan Standard | 52B (389B active) | INT4 Q8 | 12 tok/s 7 tok/s | ~4.7s ~4.8s | |
#164 | Qwen2.5-32B | 32B | INT4 Q8 FP16 | 121 tok/s 73 tok/s 37 tok/s | ~409 ms ~415 ms ~462 ms | |
#168 | Magistral Small | 24B | INT4 Q8 FP16 | 119 tok/s 81 tok/s 47 tok/s | ~311 ms ~315 ms ~324 ms | |
#169 | OLMo 3 7B Think | 7B | INT4 Q8 FP16 | 264 tok/s 204 tok/s 133 tok/s | ~96 ms ~97 ms ~99 ms | |
#170 | OLMo 3.1 32B Instruct | 32B | INT4 Q8 FP16 | 97 tok/s 64 tok/s 34 tok/s | ~411 ms ~416 ms ~464 ms | |
#171 | Llama 3.1 70B | 70B | INT4 Q8 | 62 tok/s 34 tok/s | ~885 ms ~973 ms | |
#172 | DeepSeek-R1 14B | 14B | INT4 Q8 FP16 | 226 tok/s 147 tok/s 83 tok/s | ~184 ms ~186 ms ~191 ms | |
#173 | Ministral 3 14B | 14B | INT4 Q8 FP16 | 173 tok/s 124 tok/s 75 tok/s | ~185 ms ~187 ms ~193 ms | |
#174 | Mistral-Large-2407 | 123B | INT4 | 35 tok/s | ~1.7s | |
#175 | Llama 4 Scout | 109B (17B active) | INT4 | 172 tok/s | ~261 ms | |
#177 | Gemma 3 12B | 12B | INT4 Q8 FP16 | 244 tok/s 163 tok/s 94 tok/s | ~159 ms ~161 ms ~165 ms | |
#178 | Command A | 111B | INT4 | 30 tok/s | ~1.5s | |
#179 | Qwen2-72B | 72B | INT4 Q8 | 60 tok/s 33 tok/s | ~910 ms ~1.0s | |
#180 | OLMo 3 32B Base | 32B | INT4 Q8 FP16 | 97 tok/s 64 tok/s 34 tok/s | ~411 ms ~416 ms ~464 ms | |
#181 | Mistral-Small-2501 | 24B | INT4 Q8 FP16 | 151 tok/s 94 tok/s 51 tok/s | ~309 ms ~314 ms ~322 ms | |
#182 | Ministral 3 8B | 8B | INT4 Q8 FP16 | 246 tok/s 187 tok/s 120 tok/s | ~108 ms ~110 ms ~113 ms | |
#183 | Qwen3.5-2B | 2B | INT4 Q8 FP16 | 507 tok/s 433 tok/s 327 tok/s | ~32 ms ~32 ms ~33 ms | |
#184 | MiniCPM5-2B | 2B | INT4 Q8 FP16 | 507 tok/s 433 tok/s 327 tok/s | ~32 ms ~32 ms ~33 ms | |
#185 | Qwen2.5-14B | 14B | INT4 Q8 FP16 | 221 tok/s 145 tok/s 83 tok/s | ~184 ms ~186 ms ~192 ms | |
#186 | Gemma 2 27B | 27B | INT4 Q8 FP16 | 138 tok/s 85 tok/s 46 tok/s | ~347 ms ~351 ms ~362 ms | |
#187 | OLMo 3.1 32B Think | 32B | INT4 Q8 FP16 | 97 tok/s 64 tok/s 34 tok/s | ~411 ms ~416 ms ~464 ms | |
#188 | Gemma 4 E2B | 5.1B | INT4 Q8 FP16 | 381 tok/s 287 tok/s 185 tok/s | ~71 ms ~72 ms ~74 ms | |
#189 | Qwen3-4B | 4B | INT4 Q8 FP16 | 418 tok/s 326 tok/s 219 tok/s | ~57 ms ~58 ms ~59 ms | |
#190 | DeepSeek-R1 8B | 8B | INT4 Q8 FP16 | 314 tok/s 220 tok/s 133 tok/s | ~107 ms ~109 ms ~112 ms | |
#191 | Llama 3 70B | 70B | INT4 Q8 | 62 tok/s 34 tok/s | ~885 ms ~973 ms | |
#192 | Command R Plus | 104B | INT4 | 33 tok/s | ~1.4s | |
#193 | Ministral 3 3B | 3B | INT4 Q8 FP16 | 369 tok/s 315 tok/s 233 tok/s | ~45 ms ~45 ms ~46 ms | |
#195 | Qwen3-14B | 14B | INT4 Q8 FP16 | 221 tok/s 145 tok/s 83 tok/s | ~184 ms ~186 ms ~192 ms | |
#196 | Kimi Linear 48B A3B Instruct | 48B (3B active) | INT4 Q8 | 201 tok/s 190 tok/s | ~64 ms ~64 ms | |
#197 | Gemma 2 9B | 9B | INT4 Q8 FP16 | 290 tok/s 201 tok/s 120 tok/s | ~120 ms ~122 ms ~125 ms | |
#198 | Qwen2.5-7B | 7B | INT4 Q8 FP16 | 330 tok/s 237 tok/s 147 tok/s | ~95 ms ~96 ms ~99 ms | |
#199 | Qwen3-8B | 8B | INT4 Q8 FP16 | 309 tok/s 218 tok/s 132 tok/s | ~107 ms ~109 ms ~112 ms | |
#200 | Gemma 3 4B | 4B | INT4 Q8 FP16 | 418 tok/s 326 tok/s 219 tok/s | ~57 ms ~58 ms ~59 ms | |
#201 | DeepSeek-R1 7B | 7B | INT4 Q8 FP16 | 335 tok/s 240 tok/s 147 tok/s | ~95 ms ~96 ms ~99 ms | |
#202 | Llama 3.1 8B | 8B | INT4 Q8 FP16 | 309 tok/s 218 tok/s 132 tok/s | ~107 ms ~109 ms ~112 ms | |
#204 | Qwen3-1.7B | 1.7B | INT4 Q8 FP16 | 525 tok/s 457 tok/s 354 tok/s | ~28 ms ~28 ms ~29 ms | |
#205 | Command R | 35B | INT4 Q8 FP16 | 91 tok/s 60 tok/s 30 tok/s | ~448 ms ~454 ms ~547 ms | |
#206 | Gemma 3 270M | 0.27B | INT4 Q8 FP16 | 621 tok/s 605 tok/s 571 tok/s | ~7 ms ~7 ms ~7 ms | |
#208 | Llama 3 8B | 8B | INT4 Q8 FP16 | 309 tok/s 218 tok/s 132 tok/s | ~107 ms ~109 ms ~112 ms | |
#209 | Qwen3-0.6B | 0.6B | INT4 Q8 FP16 | 609 tok/s 574 tok/s 509 tok/s | ~12 ms ~12 ms ~12 ms | |
#210 | Phi-4-Mini | 3.8B | INT4 Q8 FP16 | 425 tok/s 334 tok/s 227 tok/s | ~54 ms ~55 ms ~56 ms | |
#211 | Mixtral-8x7B-v0.1 | 46.7B (7B active) | INT4 Q8 | 305 tok/s 225 tok/s | ~110 ms ~111 ms | |
#212 | Ministral-8B-2410 | 8B | INT4 Q8 FP16 | 309 tok/s 218 tok/s 132 tok/s | ~107 ms ~109 ms ~112 ms | |
#213 | DeepSeek-R1 1.5B | 1.5B | INT4 Q8 FP16 | 538 tok/s 474 tok/s 375 tok/s | ~25 ms ~25 ms ~26 ms | |
#214 | Llama 3.2 3B | 3B | INT4 Q8 FP16 | 458 tok/s 372 tok/s 263 tok/s | ~44 ms ~45 ms ~46 ms | |
#215 | Qwen3.5-0.8B | 0.8B | INT4 Q8 FP16 | 591 tok/s 548 tok/s 471 tok/s | ~15 ms ~15 ms ~15 ms | |
#216 | Phi-3-mini | 3.8B | INT4 Q8 FP16 | 425 tok/s 334 tok/s 227 tok/s | ~54 ms ~55 ms ~56 ms | |
#217 | Gemma 3n E2B IT | 6B (2B active) | INT4 Q8 FP16 | 369 tok/s 335 tok/s 268 tok/s | ~34 ms ~35 ms ~35 ms | |
#218 | Mistral-7B-Instruct-v0.1 | 7.3B | INT4 Q8 FP16 | 324 tok/s 231 tok/s 142 tok/s | ~99 ms ~100 ms ~102 ms | |
#219 | Llama 3.2 1B | 1B | INT4 Q8 FP16 | 574 tok/s 523 tok/s 438 tok/s | ~18 ms ~18 ms ~19 ms | |
#220 | Gemma 2 2B | 2B | INT4 Q8 FP16 | 507 tok/s 433 tok/s 327 tok/s | ~32 ms ~32 ms ~33 ms | |
#221 | Phi-3-medium | 14B | INT4 Q8 FP16 | 221 tok/s 145 tok/s 83 tok/s | ~184 ms ~186 ms ~192 ms | |
#222 | OLMo 3 7B Instruct | 7B | INT4 Q8 FP16 | 264 tok/s 204 tok/s 133 tok/s | ~96 ms ~97 ms ~99 ms | |
#223 | Yi-34B | 34B | INT4 Q8 FP16 | 92 tok/s 61 tok/s 33 tok/s | ~436 ms ~442 ms ~492 ms | |
#224 | Qwen2-7B | 7B | INT4 Q8 FP16 | 330 tok/s 237 tok/s 147 tok/s | ~95 ms ~96 ms ~99 ms | |
#225 | Phi-3-small | 7B | INT4 Q8 FP16 | 330 tok/s 237 tok/s 147 tok/s | ~95 ms ~96 ms ~99 ms | |
#226 | Gemma 3 1B | 1B | INT4 Q8 FP16 | 574 tok/s 523 tok/s 438 tok/s | ~18 ms ~18 ms ~19 ms | |
#227 | Mistral-7B-Instruct-v0.2 | 7.3B | INT4 Q8 FP16 | 324 tok/s 231 tok/s 142 tok/s | ~99 ms ~100 ms ~102 ms | |
#229 | Gemma 1 7B | 7B | INT4 Q8 FP16 | 264 tok/s 204 tok/s 133 tok/s | ~96 ms ~97 ms ~99 ms | |
#230 | Gemma 1 2B | 2B | INT4 Q8 FP16 | 520 tok/s 442 tok/s 332 tok/s | ~32 ms ~32 ms ~33 ms | |
#231 | ChatGLM3-6B | 6B | INT4 Q8 FP16 | 284 tok/s 223 tok/s 149 tok/s | ~83 ms ~84 ms ~86 ms | |
#232 | ChatGLM2-6B | 6B | INT4 Q8 FP16 | 284 tok/s 223 tok/s 149 tok/s | ~83 ms ~84 ms ~86 ms | |
#233 | ChatGLM-6B | 6B | INT4 Q8 FP16 | 284 tok/s 223 tok/s 149 tok/s | ~83 ms ~84 ms ~86 ms | |
Unranked Models (Sorted by Speed) | ||||||
- | Qwen2-0.5B | 0.5B | INT4 Q8 FP16 | 619 tok/s 588 tok/s 530 tok/s | ~10 ms ~11 ms ~11 ms | |
- | Qwen2.5-0.5B | 0.5B | INT4 Q8 FP16 | 619 tok/s 588 tok/s 530 tok/s | ~10 ms ~11 ms ~11 ms | |
- | ERNIE-4.5-0.3B | 0.3B | INT4 Q8 FP16 | 615 tok/s 598 tok/s 561 tok/s | ~8 ms ~8 ms ~8 ms | |
- | ERNIE-4.5-0.3B-Base | 0.3B | INT4 Q8 FP16 | 615 tok/s 598 tok/s 561 tok/s | ~8 ms ~8 ms ~8 ms | |
- | Falcon-1B | 1B | INT4 Q8 FP16 | 584 tok/s 531 tok/s 443 tok/s | ~18 ms ~18 ms ~19 ms | |
- | Falcon3-1B | 1B | INT4 Q8 FP16 | 574 tok/s 523 tok/s 438 tok/s | ~18 ms ~18 ms ~19 ms | |
- | Qwen2-1.5B | 1.5B | INT4 Q8 FP16 | 538 tok/s 474 tok/s 375 tok/s | ~25 ms ~25 ms ~26 ms | |
- | Qwen2.5-1.5B | 1.5B | INT4 Q8 FP16 | 538 tok/s 474 tok/s 375 tok/s | ~25 ms ~25 ms ~26 ms | |
- | Spark X2.5 1.7B | 1.7B | INT4 Q8 FP16 | 525 tok/s 457 tok/s 354 tok/s | ~28 ms ~28 ms ~29 ms | |
- | Hy-MT2 1.8B | 1.8B | INT4 Q8 FP16 | 519 tok/s 449 tok/s 345 tok/s | ~29 ms ~29 ms ~30 ms | |
- | Falcon-3B | 3B | INT4 Q8 FP16 | 470 tok/s 379 tok/s 266 tok/s | ~44 ms ~45 ms ~46 ms | |
- | CroissantLLM Base | 1.3B | INT4 Q8 FP16 | 466 tok/s 428 tok/s 355 tok/s | ~22 ms ~23 ms ~23 ms | |
- | Phi-1 | 1.3B | INT4 Q8 FP16 | 466 tok/s 428 tok/s 355 tok/s | ~22 ms ~23 ms ~23 ms | |
- | Phi-1.5 | 1.3B | INT4 Q8 FP16 | 466 tok/s 428 tok/s 355 tok/s | ~22 ms ~23 ms ~23 ms | |
- | DeepSeek-R1 3B | 3B | INT4 Q8 FP16 | 461 tok/s 374 tok/s 263 tok/s | ~44 ms ~45 ms ~46 ms | |
- | Falcon3-3B | 3B | INT4 Q8 FP16 | 458 tok/s 372 tok/s 263 tok/s | ~44 ms ~45 ms ~46 ms | |
- | Qwen2.5-3B | 3B | INT4 Q8 FP16 | 458 tok/s 372 tok/s 263 tok/s | ~44 ms ~45 ms ~46 ms | |
- | ERNIE-4.5-21B-A3B | 21B (3B active) | INT4 Q8 FP16 | 428 tok/s 353 tok/s 253 tok/s | ~53 ms ~54 ms ~55 ms | |
- | ERNIE-4.5-21B-A3B-Base | 21B (3B active) | INT4 Q8 FP16 | 428 tok/s 353 tok/s 253 tok/s | ~53 ms ~54 ms ~55 ms | |
- | ERNIE-4.5-VL-28B-A3B | 28B (3B active) | INT4 Q8 FP16 | 420 tok/s 348 tok/s 251 tok/s | ~56 ms ~57 ms ~58 ms | |
- | ERNIE-4.5-VL-28B-A3B-Base | 28B (3B active) | INT4 Q8 FP16 | 420 tok/s 348 tok/s 251 tok/s | ~56 ms ~57 ms ~58 ms | |
- | Hy-MT2 30B A3B | 30B (3B active) | INT4 Q8 FP16 | 418 tok/s 347 tok/s 250 tok/s | ~57 ms ~57 ms ~58 ms | |
- | Spark X2.5 4B | 4B | INT4 Q8 FP16 | 418 tok/s 326 tok/s 219 tok/s | ~57 ms ~58 ms ~59 ms | |
- | Phi-2 | 2.7B | INT4 Q8 FP16 | 380 tok/s 329 tok/s 247 tok/s | ~41 ms ~41 ms ~43 ms | |
- | MaLLaM-3B | 3B | INT4 Q8 FP16 | 369 tok/s 315 tok/s 233 tok/s | ~45 ms ~45 ms ~46 ms | |
- | SmolLM3 3B | 3B | INT4 Q8 FP16 | 369 tok/s 315 tok/s 233 tok/s | ~45 ms ~45 ms ~46 ms | |
- | AliceAI Foundation 80B A3B Base | 80B (3B active) | INT4 Q8 | 366 tok/s 282 tok/s | ~77 ms ~90 ms | |
- | Falcon-7B | 7B | INT4 Q8 FP16 | 339 tok/s 242 tok/s 148 tok/s | ~95 ms ~96 ms ~99 ms | |
- | Falcon3-7B | 7B | INT4 Q8 FP16 | 330 tok/s 237 tok/s 147 tok/s | ~95 ms ~96 ms ~99 ms | |
- | Hy-MT2 7B | 7B | INT4 Q8 FP16 | 330 tok/s 237 tok/s 147 tok/s | ~95 ms ~96 ms ~99 ms | |
- | Mistral-7B-v0.1 | 7.3B | INT4 Q8 FP16 | 324 tok/s 231 tok/s 142 tok/s | ~99 ms ~100 ms ~102 ms | |
- | LensVLM 9B | 9B | INT4 Q8 FP16 | 290 tok/s 201 tok/s 120 tok/s | ~120 ms ~122 ms ~125 ms | |
- | MiMo V2.6 Distill Qwen 9B | 9B | INT4 Q8 FP16 | 290 tok/s 201 tok/s 120 tok/s | ~120 ms ~122 ms ~125 ms | |
- | ChatGLM3-6B-32K | 6B | INT4 Q8 FP16 | 284 tok/s 223 tok/s 149 tok/s | ~83 ms ~84 ms ~86 ms | |
- | Yi-6B | 6B | INT4 Q8 FP16 | 284 tok/s 223 tok/s 149 tok/s | ~83 ms ~84 ms ~86 ms | |
- | Falcon3-10B | 10B | INT4 Q8 FP16 | 273 tok/s 187 tok/s 110 tok/s | ~133 ms ~135 ms ~138 ms | |
- | Kimi-VL-A3B-Instruct | 16B (3B active) | INT4 Q8 FP16 | 273 tok/s 247 tok/s 194 tok/s | ~53 ms ~53 ms ~54 ms | |
- | Falcon2-11B | 11B | INT4 Q8 FP16 | 266 tok/s 177 tok/s 103 tok/s | ~146 ms ~147 ms ~151 ms | |
- | Hunyuan Lite | 7B | INT4 Q8 FP16 | 264 tok/s 204 tok/s 133 tok/s | ~96 ms ~97 ms ~99 ms | |
- | OLMo 3 7B Base | 7B | INT4 Q8 FP16 | 264 tok/s 204 tok/s 133 tok/s | ~96 ms ~97 ms ~99 ms | |
- | SEA-LION-7B | 7.1B | INT4 Q8 FP16 | 262 tok/s 202 tok/s 132 tok/s | ~97 ms ~98 ms ~101 ms | |
- | SEA-LION-7B-Instruct | 7.1B | INT4 Q8 FP16 | 262 tok/s 202 tok/s 132 tok/s | ~97 ms ~98 ms ~101 ms | |
- | Sahabat-AI-Llama3-8B-Instruct | 8B | INT4 Q8 FP16 | 246 tok/s 187 tok/s 120 tok/s | ~108 ms ~110 ms ~113 ms | |
- | Typhoon-2-8B | 8B | INT4 Q8 FP16 | 246 tok/s 187 tok/s 120 tok/s | ~108 ms ~110 ms ~113 ms | |
- | Laguna S 2.1 | 118B (8B active) | INT4 | 245 tok/s | ~157 ms | |
- | GLM-4-9B | 9B | INT4 Q8 FP16 | 230 tok/s 172 tok/s 109 tok/s | ~121 ms ~123 ms ~126 ms | |
- | GLM-4-9B-Chat | 9B | INT4 Q8 FP16 | 230 tok/s 172 tok/s 109 tok/s | ~121 ms ~123 ms ~126 ms | |
- | GLM-4-9B-Chat-1M | 9B | INT4 Q8 FP16 | 230 tok/s 172 tok/s 109 tok/s | ~121 ms ~123 ms ~126 ms | |
- | Yi-9B | 9B | INT4 Q8 FP16 | 230 tok/s 172 tok/s 109 tok/s | ~121 ms ~123 ms ~126 ms | |
- | Sahabat-AI-Gemma2-9B | 9.2B | INT4 Q8 FP16 | 227 tok/s 170 tok/s 107 tok/s | ~124 ms ~125 ms ~129 ms | |
- | Sahabat-AI-Gemma2-9B-Instruct | 9.2B | INT4 Q8 FP16 | 227 tok/s 170 tok/s 107 tok/s | ~124 ms ~125 ms ~129 ms | |
- | GreenMind-14B-R1 | 14B | INT4 Q8 FP16 | 173 tok/s 124 tok/s 75 tok/s | ~185 ms ~187 ms ~193 ms | |
- | Hunyuan A13B | 80B (13B active) | INT4 | 110 tok/s | ~202 ms | |
- | Falcon-40B | 40B | INT4 Q8 FP16 | 103 tok/s 61 tok/s 29 tok/s | ~508 ms ~515 ms ~620 ms | |
- | Typhoon-2-70B | 70B | INT4 Q8 | 50 tok/s 28 tok/s | ~889 ms ~1.1s | |
- | Kimi-Dev-72B | 72B | INT4 Q8 | 49 tok/s 28 tok/s | ~914 ms ~1.1s | |
Inference Capacity
Maximum model parameter class runnable by weight precision.
Context:
4-bit
70B+ Models
8-bit
70B Models
16-bit
32B Models
Fine-Tuning Capacity
Hardware capacity for local training and adapter fine-tuning.
Context Length:
QLoRA
70B+ Models
LoRA
70B Models
Full Parameter Training
7B-8B Models
Workload Recommendations
Guidance for optimal precision and operational ceilings on NVIDIA H100 (80GB).
Inference Sweet Spot
Suitable for 70B parameter models at Q4/Q8 with extended 32k context windows, or high-throughput batching of smaller dense architectures.
Bandwidth & Speed Profile
With 3350 GB/s aggregate bandwidth, batch size 1 inference operates in a memory-bandwidth bound regime. Generates approximately 744 tok/s on an 8B Q4 model and 88 tok/s on a 70B Q4 model.
Frequently Asked Questions
©2025 ApX Machine Learning