NVIDIA B200 (192GB)
The NVIDIA B200 (192GB) features 192 GB of dedicated VRAM and 8000 GB/s memory bandwidth, supporting CUDA and TensorRT-LLM runtimes for full-precision and quantized inference.
Memory
192 GB
Usable Memory Ceiling
192 GB
Bandwidth
8000 GB/s
Max Inference Class
70B+ Models (Q4)
TDP
1200W
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 B200 (192GB)
Inference throughput, memory compatibility, and generation speeds for open-weights LLMs on NVIDIA B200 (192GB).
Sort by:
Context:
KV Cache:
GPUs:
| Rank | Model | Parameters | Precision | TPS (Approx.) | TTFT | Can Run |
|---|---|---|---|---|---|---|
#30 | DeepSeek-V4-Flash-Vision-Exp | 290.9B (20.4B active) | INT4 | 308 tok/s | ~114 ms | |
#33 | Qwen 3.8 27B | 27B | INT4 Q8 FP16 | 256 tok/s 172 tok/s 100 tok/s | ~156 ms ~158 ms ~162 ms | |
#40 | MiMo-V2.6-Flash | 309B | INT4 | 33 tok/s | ~1.9s | |
#41 | Qwen 3.8 Flash | 176B (6B active) | INT4 Q8 | 495 tok/s 388 tok/s | ~36 ms ~40 ms | |
#42 | DeepSeek-V4-Flash | 284B (13B active) | INT4 | 382 tok/s | ~74 ms | |
#43 | K2 Horizon 375B A23B | 375B (23B active) | INT4 | 209 tok/s | ~211 ms | |
#48 | Agnes 3.0 Flash | 33B | INT4 Q8 FP16 | 226 tok/s 148 tok/s 84 tok/s | ~189 ms ~191 ms ~196 ms | |
#61 | Sarvam-105B | 106B (10.3B active) | INT4 Q8 | 408 tok/s 315 tok/s | ~81 ms ~82 ms | |
#72 | K2 Horizon MoVA 36B A4B | 36B (4B active) | INT4 Q8 FP16 | 508 tok/s 445 tok/s 348 tok/s | ~33 ms ~33 ms ~34 ms | |
#75 | Ling 3.0 Flash VL | 124B (5.5B active) | INT4 Q8 | 489 tok/s 411 tok/s | ~58 ms ~59 ms | |
#80 | GLM-5.3 Flash | 313.3B (17.3B active) | INT4 | 318 tok/s | ~105 ms | |
#82 | Qwen3 235B A22B Thinking | 235B (22B active) | INT4 | 129 tok/s | ~160 ms | |
#83 | Qwen3.5-27B | 27B | INT4 Q8 FP16 | 256 tok/s 172 tok/s 100 tok/s | ~156 ms ~158 ms ~162 ms | |
#84 | Qwen3.5-122B-A10B | 122B (10B active) | INT4 Q8 | 362 tok/s 291 tok/s | ~83 ms ~84 ms | |
#92 | Sarvam-30B | 32B (2.4B active) | INT4 Q8 FP16 | 547 tok/s 503 tok/s 423 tok/s | ~24 ms ~24 ms ~25 ms | |
#94 | Qwen3.6 35B A3B | 35B (3B active) | INT4 Q8 FP16 | 531 tok/s 479 tok/s 391 tok/s | ~28 ms ~28 ms ~28 ms | |
#96 | Muse Glimmer 30B | 30B | INT4 Q8 FP16 | 240 tok/s 159 tok/s 91 tok/s | ~173 ms ~175 ms ~179 ms | |
#100 | Phi-4 Reasoning Plus | 14B | INT4 Q8 FP16 | 304 tok/s 236 tok/s 156 tok/s | ~84 ms ~85 ms ~87 ms | |
#101 | MiMo V2 Flash | 15B (309B active) | INT4 Q8 FP16 | 35 tok/s 20 tok/s 10 tok/s | ~1.7s ~1.7s ~1.7s | |
#105 | Inkling-Small | 276B (12B active) | INT4 | 122 tok/s | ~123 ms | |
#106 | GLM-4.5-Air | 106B (12B active) | INT4 Q8 | 186 tok/s 167 tok/s | ~93 ms ~93 ms | |
#108 | Gemma 4 26B A4B | 25.2B (3.8B active) | INT4 Q8 FP16 | 518 tok/s 456 tok/s 359 tok/s | ~31 ms ~31 ms ~31 ms | |
#109 | Gemma 4 12B | 11.95B | INT4 Q8 FP16 | 328 tok/s 259 tok/s 175 tok/s | ~73 ms ~73 ms ~75 ms | |
#111 | MiniMax M2 | 229B (10B active) | INT4 | 150 tok/s | ~95 ms | |
#113 | Qwen3.5-9B | 9B | INT4 Q8 FP16 | 431 tok/s 338 tok/s 229 tok/s | ~55 ms ~56 ms ~57 ms | |
#114 | Qwen3.5-35B-A3B | 35B (3B active) | INT4 Q8 FP16 | 531 tok/s 479 tok/s 391 tok/s | ~28 ms ~28 ms ~28 ms | |
#114 | Ling 3.0 Flash | 124B (5.1B active) | INT4 Q8 | 498 tok/s 422 tok/s | ~56 ms ~57 ms | |
#118 | K2 Horizon 7B | 7B | INT4 Q8 FP16 | 467 tok/s 379 tok/s 268 tok/s | ~44 ms ~45 ms ~46 ms | |
#119 | Step 3.5 Flash | 196.81B (11B active) | INT4 Q8 | 345 tok/s 247 tok/s | ~91 ms ~108 ms | |
#126 | Qwen3 Next 80B A3B | 80B (79B active) | INT4 Q8 FP16 | 97 tok/s 63 tok/s 34 tok/s | ~447 ms ~453 ms ~503 ms | |
#128 | NVIDIA Nemotron 3 Nano 30B-A3B | 3.5B (30B active) | INT4 Q8 FP16 | 232 tok/s 155 tok/s 90 tok/s | ~167 ms ~169 ms ~174 ms | |
#129 | Gemma 4 31B | 30.7B | INT4 Q8 FP16 | 237 tok/s 156 tok/s 89 tok/s | ~176 ms ~178 ms ~183 ms | |
#130 | GPT-OSS 120B | 117B (5.1B active) | INT4 Q8 | 195 tok/s 191 tok/s | ~59 ms ~59 ms | |
#131 | Nemotron 3.5 Lightning | 30B (3B active) | INT4 Q8 FP16 | 369 tok/s 351 tok/s 301 tok/s | ~28 ms ~28 ms ~29 ms | |
#132 | Qwen3.5-4B | 4B | INT4 Q8 FP16 | 533 tok/s 464 tok/s 359 tok/s | ~27 ms ~28 ms ~28 ms | |
#133 | Qwen3-30B-A3B | 30B (3B active) | INT4 Q8 FP16 | 533 tok/s 481 tok/s 393 tok/s | ~27 ms ~27 ms ~28 ms | |
#136 | GPT-OSS 20B | 21B (3.6B active) | INT4 Q8 FP16 | 389 tok/s 362 tok/s 300 tok/s | ~30 ms ~30 ms ~30 ms | |
#139 | Qwen3-235B-A22B | 235B (22B active) | INT4 | 255 tok/s | ~156 ms | |
#144 | Mistral Medium 3.5 128B | 128B | INT4 Q8 | 64 tok/s 41 tok/s | ~721 ms ~730 ms | |
#146 | Granite 4.2 8B | 8B | INT4 Q8 FP16 | 449 tok/s 358 tok/s 247 tok/s | ~50 ms ~50 ms ~51 ms | |
#150 | Gemma 4 E4B | 8B | INT4 Q8 FP16 | 449 tok/s 358 tok/s 247 tok/s | ~50 ms ~50 ms ~51 ms | |
#151 | Nemotron 3.5 Lightning 30B A3B | 30B (3B active) | INT4 Q8 FP16 | 533 tok/s 481 tok/s 393 tok/s | ~27 ms ~27 ms ~28 ms | |
#153 | Qwen3-32B | 32B | INT4 Q8 FP16 | 230 tok/s 151 tok/s 86 tok/s | ~183 ms ~186 ms ~191 ms | |
#154 | Qwen2.5-72B | 72B | INT4 Q8 FP16 | 127 tok/s 77 tok/s 41 tok/s | ~406 ms ~411 ms ~422 ms | |
#155 | Mistral Large 3 | 41B (675B active) | INT4 Q8 FP16 | 17 tok/s 9 tok/s 5 tok/s | ~3.6s ~3.7s ~3.8s | |
#156 | GLM-4 | 32B | INT4 Q8 FP16 | 192 tok/s 135 tok/s 81 tok/s | ~184 ms ~187 ms ~192 ms | |
#157 | DeepSeek-R1 70B | 70B | INT4 Q8 FP16 | 133 tok/s 80 tok/s 43 tok/s | ~395 ms ~400 ms ~410 ms | |
#158 | GLM-4V | 9B | INT4 Q8 FP16 | 371 tok/s 303 tok/s 212 tok/s | ~56 ms ~56 ms ~58 ms | |
#159 | Gemma 3 27B | 27B | INT4 Q8 FP16 | 256 tok/s 172 tok/s 100 tok/s | ~156 ms ~158 ms ~162 ms | |
#160 | Phi-4 | 14B | INT4 Q8 FP16 | 362 tok/s 266 tok/s 168 tok/s | ~84 ms ~85 ms ~87 ms | |
#161 | Llama 3.3 70B | 70B | INT4 Q8 FP16 | 130 tok/s 79 tok/s 42 tok/s | ~395 ms ~400 ms ~411 ms | |
#162 | DeepSeek-R1 32B | 32B | INT4 Q8 FP16 | 235 tok/s 153 tok/s 87 tok/s | ~183 ms ~186 ms ~191 ms | |
#163 | Hunyuan Standard | 52B (389B active) | INT4 Q8 FP16 | 28 tok/s 16 tok/s 8 tok/s | ~2.1s ~2.1s ~2.2s | |
#164 | Qwen2.5-32B | 32B | INT4 Q8 FP16 | 230 tok/s 151 tok/s 86 tok/s | ~183 ms ~186 ms ~191 ms | |
#168 | Magistral Small | 24B | INT4 Q8 FP16 | 228 tok/s 166 tok/s 102 tok/s | ~140 ms ~142 ms ~145 ms | |
#169 | OLMo 3 7B Think | 7B | INT4 Q8 FP16 | 407 tok/s 341 tok/s 249 tok/s | ~44 ms ~45 ms ~46 ms | |
#170 | OLMo 3.1 32B Instruct | 32B | INT4 Q8 FP16 | 192 tok/s 135 tok/s 81 tok/s | ~184 ms ~187 ms ~192 ms | |
#171 | Llama 3.1 70B | 70B | INT4 Q8 FP16 | 130 tok/s 79 tok/s 42 tok/s | ~395 ms ~400 ms ~411 ms | |
#172 | DeepSeek-R1 14B | 14B | INT4 Q8 FP16 | 367 tok/s 269 tok/s 169 tok/s | ~84 ms ~85 ms ~87 ms | |
#173 | Ministral 3 14B | 14B | INT4 Q8 FP16 | 304 tok/s 236 tok/s 156 tok/s | ~84 ms ~85 ms ~87 ms | |
#174 | Mistral-Large-2407 | 123B | INT4 Q8 | 81 tok/s 47 tok/s | ~692 ms ~700 ms | |
#175 | Llama 4 Scout | 109B (17B active) | INT4 Q8 | 302 tok/s 223 tok/s | ~118 ms ~119 ms | |
#177 | Gemma 3 12B | 12B | INT4 Q8 FP16 | 387 tok/s 291 tok/s 188 tok/s | ~72 ms ~73 ms ~75 ms | |
#178 | Command A | 111B | INT4 Q8 | 72 tok/s 46 tok/s | ~627 ms ~635 ms | |
#179 | Qwen2-72B | 72B | INT4 Q8 FP16 | 127 tok/s 77 tok/s 41 tok/s | ~406 ms ~411 ms ~422 ms | |
#180 | OLMo 3 32B Base | 32B | INT4 Q8 FP16 | 192 tok/s 135 tok/s 81 tok/s | ~184 ms ~187 ms ~192 ms | |
#181 | Mistral-Small-2501 | 24B | INT4 Q8 FP16 | 275 tok/s 187 tok/s 110 tok/s | ~139 ms ~141 ms ~145 ms | |
#182 | Ministral 3 8B | 8B | INT4 Q8 FP16 | 389 tok/s 321 tok/s 229 tok/s | ~50 ms ~51 ms ~52 ms | |
#183 | Qwen3.5-2B | 2B | INT4 Q8 FP16 | 589 tok/s 544 tok/s 465 tok/s | ~16 ms ~16 ms ~17 ms | |
#184 | MiniCPM5-2B | 2B | INT4 Q8 FP16 | 589 tok/s 544 tok/s 465 tok/s | ~16 ms ~16 ms ~17 ms | |
#185 | Qwen2.5-14B | 14B | INT4 Q8 FP16 | 362 tok/s 266 tok/s 168 tok/s | ~84 ms ~85 ms ~87 ms | |
#186 | Gemma 2 27B | 27B | INT4 Q8 FP16 | 256 tok/s 172 tok/s 100 tok/s | ~156 ms ~158 ms ~162 ms | |
#187 | OLMo 3.1 32B Think | 32B | INT4 Q8 FP16 | 192 tok/s 135 tok/s 81 tok/s | ~184 ms ~187 ms ~192 ms | |
#188 | Gemma 4 E2B | 5.1B | INT4 Q8 FP16 | 507 tok/s 429 tok/s 320 tok/s | ~33 ms ~34 ms ~35 ms | |
#189 | Qwen3-4B | 4B | INT4 Q8 FP16 | 533 tok/s 464 tok/s 359 tok/s | ~27 ms ~28 ms ~28 ms | |
#190 | DeepSeek-R1 8B | 8B | INT4 Q8 FP16 | 453 tok/s 360 tok/s 248 tok/s | ~50 ms ~50 ms ~51 ms | |
#191 | Llama 3 70B | 70B | INT4 Q8 FP16 | 130 tok/s 79 tok/s 42 tok/s | ~395 ms ~400 ms ~411 ms | |
#192 | Command R Plus | 104B | INT4 Q8 | 76 tok/s 49 tok/s | ~588 ms ~595 ms | |
#193 | Ministral 3 3B | 3B | INT4 Q8 FP16 | 499 tok/s 455 tok/s 375 tok/s | ~22 ms ~22 ms ~23 ms | |
#194 | ERNIE-4.5-300B-A47B | 300B (47B active) | INT4 | 153 tok/s | ~327 ms | |
#195 | Qwen3-14B | 14B | INT4 Q8 FP16 | 362 tok/s 266 tok/s 168 tok/s | ~84 ms ~85 ms ~87 ms | |
#196 | Kimi Linear 48B A3B Instruct | 48B (3B active) | INT4 Q8 FP16 | 339 tok/s 326 tok/s 282 tok/s | ~30 ms ~30 ms ~31 ms | |
#197 | Gemma 2 9B | 9B | INT4 Q8 FP16 | 431 tok/s 338 tok/s 229 tok/s | ~55 ms ~56 ms ~57 ms | |
#198 | Qwen2.5-7B | 7B | INT4 Q8 FP16 | 467 tok/s 379 tok/s 268 tok/s | ~44 ms ~45 ms ~46 ms | |
#199 | Qwen3-8B | 8B | INT4 Q8 FP16 | 449 tok/s 358 tok/s 247 tok/s | ~50 ms ~50 ms ~51 ms | |
#200 | Gemma 3 4B | 4B | INT4 Q8 FP16 | 533 tok/s 464 tok/s 359 tok/s | ~27 ms ~28 ms ~28 ms | |
#201 | DeepSeek-R1 7B | 7B | INT4 Q8 FP16 | 472 tok/s 382 tok/s 269 tok/s | ~44 ms ~45 ms ~46 ms | |
#202 | Llama 3.1 8B | 8B | INT4 Q8 FP16 | 449 tok/s 358 tok/s 247 tok/s | ~50 ms ~50 ms ~51 ms | |
#203 | Mixtral-8x22B-v0.1 | 176B (22B active) | INT4 Q8 | 265 tok/s 179 tok/s | ~148 ms ~162 ms | |
#204 | Qwen3-1.7B | 1.7B | INT4 Q8 FP16 | 599 tok/s 559 tok/s 487 tok/s | ~14 ms ~14 ms ~15 ms | |
#205 | Command R | 35B | INT4 Q8 FP16 | 182 tok/s 127 tok/s 75 tok/s | ~201 ms ~203 ms ~209 ms | |
#206 | Gemma 3 270M | 0.27B | INT4 Q8 FP16 | 647 tok/s 640 tok/s 623 tok/s | ~5 ms ~5 ms ~5 ms | |
#208 | Llama 3 8B | 8B | INT4 Q8 FP16 | 449 tok/s 358 tok/s 247 tok/s | ~50 ms ~50 ms ~51 ms | |
#209 | Qwen3-0.6B | 0.6B | INT4 Q8 FP16 | 641 tok/s 625 tok/s 590 tok/s | ~7 ms ~7 ms ~7 ms | |
#210 | Phi-4-Mini | 3.8B | INT4 Q8 FP16 | 539 tok/s 471 tok/s 368 tok/s | ~26 ms ~26 ms ~27 ms | |
#211 | Mixtral-8x7B-v0.1 | 46.7B (7B active) | INT4 Q8 FP16 | 446 tok/s 366 tok/s 261 tok/s | ~50 ms ~51 ms ~52 ms | |
#212 | Ministral-8B-2410 | 8B | INT4 Q8 FP16 | 449 tok/s 358 tok/s 247 tok/s | ~50 ms ~50 ms ~51 ms | |
#213 | DeepSeek-R1 1.5B | 1.5B | INT4 Q8 FP16 | 606 tok/s 570 tok/s 503 tok/s | ~13 ms ~13 ms ~13 ms | |
#214 | Llama 3.2 3B | 3B | INT4 Q8 FP16 | 560 tok/s 501 tok/s 405 tok/s | ~22 ms ~22 ms ~22 ms | |
#215 | Qwen3.5-0.8B | 0.8B | INT4 Q8 FP16 | 633 tok/s 611 tok/s 568 tok/s | ~9 ms ~9 ms ~9 ms | |
#216 | Phi-3-mini | 3.8B | INT4 Q8 FP16 | 539 tok/s 471 tok/s 368 tok/s | ~26 ms ~26 ms ~27 ms | |
#217 | Gemma 3n E2B IT | 6B (2B active) | INT4 Q8 FP16 | 498 tok/s 471 tok/s 411 tok/s | ~17 ms ~17 ms ~18 ms | |
#218 | Mistral-7B-Instruct-v0.1 | 7.3B | INT4 Q8 FP16 | 462 tok/s 373 tok/s 262 tok/s | ~46 ms ~46 ms ~47 ms | |
#219 | Llama 3.2 1B | 1B | INT4 Q8 FP16 | 624 tok/s 598 tok/s 547 tok/s | ~10 ms ~10 ms ~10 ms | |
#220 | Gemma 2 2B | 2B | INT4 Q8 FP16 | 589 tok/s 544 tok/s 465 tok/s | ~16 ms ~16 ms ~17 ms | |
#221 | Phi-3-medium | 14B | INT4 Q8 FP16 | 362 tok/s 266 tok/s 168 tok/s | ~84 ms ~85 ms ~87 ms | |
#222 | OLMo 3 7B Instruct | 7B | INT4 Q8 FP16 | 407 tok/s 341 tok/s 249 tok/s | ~44 ms ~45 ms ~46 ms | |
#223 | Yi-34B | 34B | INT4 Q8 FP16 | 185 tok/s 129 tok/s 77 tok/s | ~195 ms ~198 ms ~203 ms | |
#224 | Qwen2-7B | 7B | INT4 Q8 FP16 | 467 tok/s 379 tok/s 268 tok/s | ~44 ms ~45 ms ~46 ms | |
#225 | Phi-3-small | 7B | INT4 Q8 FP16 | 467 tok/s 379 tok/s 268 tok/s | ~44 ms ~45 ms ~46 ms | |
#226 | Gemma 3 1B | 1B | INT4 Q8 FP16 | 624 tok/s 598 tok/s 547 tok/s | ~10 ms ~10 ms ~10 ms | |
#227 | Mistral-7B-Instruct-v0.2 | 7.3B | INT4 Q8 FP16 | 462 tok/s 373 tok/s 262 tok/s | ~46 ms ~46 ms ~47 ms | |
#228 | Falcon-180B | 180B | INT4 Q8 | 59 tok/s 32 tok/s | ~999 ms ~1.1s | |
#229 | Gemma 1 7B | 7B | INT4 Q8 FP16 | 407 tok/s 341 tok/s 249 tok/s | ~44 ms ~45 ms ~46 ms | |
#230 | Gemma 1 2B | 2B | INT4 Q8 FP16 | 596 tok/s 550 tok/s 469 tok/s | ~16 ms ~16 ms ~16 ms | |
#231 | ChatGLM3-6B | 6B | INT4 Q8 FP16 | 426 tok/s 364 tok/s 271 tok/s | ~39 ms ~39 ms ~40 ms | |
#232 | ChatGLM2-6B | 6B | INT4 Q8 FP16 | 426 tok/s 364 tok/s 271 tok/s | ~39 ms ~39 ms ~40 ms | |
#233 | ChatGLM-6B | 6B | INT4 Q8 FP16 | 426 tok/s 364 tok/s 271 tok/s | ~39 ms ~39 ms ~40 ms | |
Unranked Models (Sorted by Speed) | ||||||
- | Qwen2-0.5B | 0.5B | INT4 Q8 FP16 | 646 tok/s 631 tok/s 602 tok/s | ~7 ms ~7 ms ~7 ms | |
- | Qwen2.5-0.5B | 0.5B | INT4 Q8 FP16 | 646 tok/s 631 tok/s 602 tok/s | ~7 ms ~7 ms ~7 ms | |
- | ERNIE-4.5-0.3B | 0.3B | INT4 Q8 FP16 | 644 tok/s 636 tok/s 618 tok/s | ~5 ms ~5 ms ~5 ms | |
- | ERNIE-4.5-0.3B-Base | 0.3B | INT4 Q8 FP16 | 644 tok/s 636 tok/s 618 tok/s | ~5 ms ~5 ms ~5 ms | |
- | Falcon-1B | 1B | INT4 Q8 FP16 | 629 tok/s 602 tok/s 551 tok/s | ~10 ms ~10 ms ~10 ms | |
- | Falcon3-1B | 1B | INT4 Q8 FP16 | 624 tok/s 598 tok/s 547 tok/s | ~10 ms ~10 ms ~10 ms | |
- | Qwen2-1.5B | 1.5B | INT4 Q8 FP16 | 606 tok/s 570 tok/s 503 tok/s | ~13 ms ~13 ms ~13 ms | |
- | Qwen2.5-1.5B | 1.5B | INT4 Q8 FP16 | 606 tok/s 570 tok/s 503 tok/s | ~13 ms ~13 ms ~13 ms | |
- | Spark X2.5 1.7B | 1.7B | INT4 Q8 FP16 | 599 tok/s 559 tok/s 487 tok/s | ~14 ms ~14 ms ~15 ms | |
- | Hy-MT2 1.8B | 1.8B | INT4 Q8 FP16 | 596 tok/s 554 tok/s 479 tok/s | ~15 ms ~15 ms ~15 ms | |
- | Falcon-3B | 3B | INT4 Q8 FP16 | 567 tok/s 506 tok/s 409 tok/s | ~22 ms ~22 ms ~22 ms | |
- | CroissantLLM Base | 1.3B | INT4 Q8 FP16 | 565 tok/s 541 tok/s 487 tok/s | ~12 ms ~12 ms ~12 ms | |
- | Phi-1 | 1.3B | INT4 Q8 FP16 | 565 tok/s 541 tok/s 487 tok/s | ~12 ms ~12 ms ~12 ms | |
- | Phi-1.5 | 1.3B | INT4 Q8 FP16 | 565 tok/s 541 tok/s 487 tok/s | ~12 ms ~12 ms ~12 ms | |
- | DeepSeek-R1 3B | 3B | INT4 Q8 FP16 | 562 tok/s 502 tok/s 406 tok/s | ~22 ms ~22 ms ~22 ms | |
- | Falcon3-3B | 3B | INT4 Q8 FP16 | 560 tok/s 501 tok/s 405 tok/s | ~22 ms ~22 ms ~22 ms | |
- | Qwen2.5-3B | 3B | INT4 Q8 FP16 | 560 tok/s 501 tok/s 405 tok/s | ~22 ms ~22 ms ~22 ms | |
- | ERNIE-4.5-21B-A3B | 21B (3B active) | INT4 Q8 FP16 | 540 tok/s 486 tok/s 396 tok/s | ~26 ms ~26 ms ~26 ms | |
- | ERNIE-4.5-21B-A3B-Base | 21B (3B active) | INT4 Q8 FP16 | 540 tok/s 486 tok/s 396 tok/s | ~26 ms ~26 ms ~26 ms | |
- | ERNIE-4.5-VL-28B-A3B | 28B (3B active) | INT4 Q8 FP16 | 535 tok/s 482 tok/s 393 tok/s | ~27 ms ~27 ms ~27 ms | |
- | ERNIE-4.5-VL-28B-A3B-Base | 28B (3B active) | INT4 Q8 FP16 | 535 tok/s 482 tok/s 393 tok/s | ~27 ms ~27 ms ~27 ms | |
- | Hy-MT2 30B A3B | 30B (3B active) | INT4 Q8 FP16 | 533 tok/s 481 tok/s 393 tok/s | ~27 ms ~27 ms ~28 ms | |
- | Spark X2.5 4B | 4B | INT4 Q8 FP16 | 533 tok/s 464 tok/s 359 tok/s | ~27 ms ~28 ms ~28 ms | |
- | Phi-2 | 2.7B | INT4 Q8 FP16 | 507 tok/s 466 tok/s 389 tok/s | ~20 ms ~20 ms ~21 ms | |
- | MaLLaM-3B | 3B | INT4 Q8 FP16 | 499 tok/s 455 tok/s 375 tok/s | ~22 ms ~22 ms ~23 ms | |
- | SmolLM3 3B | 3B | INT4 Q8 FP16 | 499 tok/s 455 tok/s 375 tok/s | ~22 ms ~22 ms ~23 ms | |
- | AliceAI Foundation 80B A3B Base | 80B (3B active) | INT4 Q8 FP16 | 496 tok/s 453 tok/s 355 tok/s | ~36 ms ~36 ms ~39 ms | |
- | Falcon-7B | 7B | INT4 Q8 FP16 | 475 tok/s 384 tok/s 270 tok/s | ~44 ms ~45 ms ~46 ms | |
- | Falcon3-7B | 7B | INT4 Q8 FP16 | 467 tok/s 379 tok/s 268 tok/s | ~44 ms ~45 ms ~46 ms | |
- | Hy-MT2 7B | 7B | INT4 Q8 FP16 | 467 tok/s 379 tok/s 268 tok/s | ~44 ms ~45 ms ~46 ms | |
- | Mistral-7B-v0.1 | 7.3B | INT4 Q8 FP16 | 462 tok/s 373 tok/s 262 tok/s | ~46 ms ~46 ms ~47 ms | |
- | LensVLM 9B | 9B | INT4 Q8 FP16 | 431 tok/s 338 tok/s 229 tok/s | ~55 ms ~56 ms ~57 ms | |
- | MiMo V2.6 Distill Qwen 9B | 9B | INT4 Q8 FP16 | 431 tok/s 338 tok/s 229 tok/s | ~55 ms ~56 ms ~57 ms | |
- | ChatGLM3-6B-32K | 6B | INT4 Q8 FP16 | 426 tok/s 364 tok/s 271 tok/s | ~39 ms ~39 ms ~40 ms | |
- | Yi-6B | 6B | INT4 Q8 FP16 | 426 tok/s 364 tok/s 271 tok/s | ~39 ms ~39 ms ~40 ms | |
- | Kimi-VL-A3B-Instruct | 16B (3B active) | INT4 Q8 FP16 | 416 tok/s 390 tok/s 329 tok/s | ~25 ms ~25 ms ~26 ms | |
- | Falcon3-10B | 10B | INT4 Q8 FP16 | 415 tok/s 321 tok/s 214 tok/s | ~61 ms ~62 ms ~63 ms | |
- | Falcon2-11B | 11B | INT4 Q8 FP16 | 408 tok/s 309 tok/s 202 tok/s | ~66 ms ~67 ms ~69 ms | |
- | Hunyuan Lite | 7B | INT4 Q8 FP16 | 407 tok/s 341 tok/s 249 tok/s | ~44 ms ~45 ms ~46 ms | |
- | OLMo 3 7B Base | 7B | INT4 Q8 FP16 | 407 tok/s 341 tok/s 249 tok/s | ~44 ms ~45 ms ~46 ms | |
- | SEA-LION-7B | 7.1B | INT4 Q8 FP16 | 405 tok/s 339 tok/s 247 tok/s | ~45 ms ~45 ms ~47 ms | |
- | SEA-LION-7B-Instruct | 7.1B | INT4 Q8 FP16 | 405 tok/s 339 tok/s 247 tok/s | ~45 ms ~45 ms ~47 ms | |
- | Sahabat-AI-Llama3-8B-Instruct | 8B | INT4 Q8 FP16 | 389 tok/s 321 tok/s 229 tok/s | ~50 ms ~51 ms ~52 ms | |
- | Typhoon-2-8B | 8B | INT4 Q8 FP16 | 389 tok/s 321 tok/s 229 tok/s | ~50 ms ~51 ms ~52 ms | |
- | Laguna S 2.1 | 118B (8B active) | INT4 Q8 | 388 tok/s 321 tok/s | ~72 ms ~72 ms | |
- | GLM-4-9B | 9B | INT4 Q8 FP16 | 371 tok/s 303 tok/s 212 tok/s | ~56 ms ~56 ms ~58 ms | |
- | GLM-4-9B-Chat | 9B | INT4 Q8 FP16 | 371 tok/s 303 tok/s 212 tok/s | ~56 ms ~56 ms ~58 ms | |
- | GLM-4-9B-Chat-1M | 9B | INT4 Q8 FP16 | 371 tok/s 303 tok/s 212 tok/s | ~56 ms ~56 ms ~58 ms | |
- | Yi-9B | 9B | INT4 Q8 FP16 | 371 tok/s 303 tok/s 212 tok/s | ~56 ms ~56 ms ~58 ms | |
- | Sahabat-AI-Gemma2-9B | 9.2B | INT4 Q8 FP16 | 368 tok/s 300 tok/s 209 tok/s | ~57 ms ~57 ms ~59 ms | |
- | Sahabat-AI-Gemma2-9B-Instruct | 9.2B | INT4 Q8 FP16 | 368 tok/s 300 tok/s 209 tok/s | ~57 ms ~57 ms ~59 ms | |
- | GreenMind-14B-R1 | 14B | INT4 Q8 FP16 | 304 tok/s 236 tok/s 156 tok/s | ~84 ms ~85 ms ~87 ms | |
- | Hunyuan A13B | 80B (13B active) | INT4 Q8 FP16 | 213 tok/s 184 tok/s 127 tok/s | ~92 ms ~92 ms ~102 ms | |
- | Falcon-40B | 40B | INT4 Q8 FP16 | 203 tok/s 129 tok/s 71 tok/s | ~227 ms ~230 ms ~236 ms | |
- | ERNIE-4.5-300B-A47B-Base | 300B (47B active) | INT4 | 153 tok/s | ~327 ms | |
- | Typhoon-2-70B | 70B | INT4 Q8 FP16 | 108 tok/s 71 tok/s 40 tok/s | ~396 ms ~401 ms ~412 ms | |
- | Kimi-Dev-72B | 72B | INT4 Q8 FP16 | 105 tok/s 69 tok/s 37 tok/s | ~408 ms ~413 ms ~459 ms | |
- | MiMo V2.6 Flash RL | 309B (15B active) | INT4 | 103 tok/s | ~159 ms | |
- | GLM-130B | 130B | INT4 Q8 | 63 tok/s 40 tok/s | ~732 ms ~741 ms | |
Inference Capacity
Maximum model parameter class runnable by weight precision.
Context:
4-bit
70B+ Models
8-bit
70B Models
16-bit
70B 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 B200 (192GB).
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 8000 GB/s aggregate bandwidth, batch size 1 inference operates in a memory-bandwidth bound regime. Generates approximately 1778 tok/s on an 8B Q4 model and 211 tok/s on a 70B Q4 model.
Frequently Asked Questions
©2025 ApX Machine Learning