Instinct MI300X (192GB)
The Instinct MI300X (192GB) provides 192 GB of memory with 5300 GB/s bandwidth, supporting local inference via ROCm and Vulkan acceleration.
Memory
192 GB
Usable Memory Ceiling
192 GB
Bandwidth
5300 GB/s
Max Inference Class
70B+ Models (Q4)
TDP
750W
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 Instinct MI300X (192GB)
Inference throughput, memory compatibility, and generation speeds for open-weights LLMs on Instinct MI300X (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 | 242 tok/s | ~193 ms | |
#33 | Qwen 3.8 27B | 27B | INT4 Q8 FP16 | 195 tok/s 125 tok/s 69 tok/s | ~265 ms ~268 ms ~274 ms | |
#40 | MiMo-V2.6-Flash | 309B | INT4 | 22 tok/s | ~3.2s | |
#41 | Qwen 3.8 Flash | 176B (6B active) | INT4 Q8 | 437 tok/s 324 tok/s | ~60 ms ~65 ms | |
#42 | DeepSeek-V4-Flash | 284B (13B active) | INT4 | 313 tok/s | ~125 ms | |
#43 | K2 Horizon 375B A23B | 375B (23B active) | INT4 | 157 tok/s | ~361 ms | |
#48 | Agnes 3.0 Flash | 33B | INT4 Q8 FP16 | 169 tok/s 106 tok/s 58 tok/s | ~321 ms ~325 ms ~333 ms | |
#61 | Sarvam-105B | 106B (10.3B active) | INT4 Q8 | 341 tok/s 248 tok/s | ~136 ms ~138 ms | |
#72 | K2 Horizon MoVA 36B A4B | 36B (4B active) | INT4 Q8 FP16 | 453 tok/s 381 tok/s 280 tok/s | ~54 ms ~55 ms ~56 ms | |
#75 | Ling 3.0 Flash VL | 124B (5.5B active) | INT4 Q8 | 431 tok/s 344 tok/s | ~98 ms ~98 ms | |
#80 | GLM-5.3 Flash | 313.3B (17.3B active) | INT4 | 254 tok/s | ~178 ms | |
#82 | Qwen3 235B A22B Thinking | 235B (22B active) | INT4 | 91 tok/s | ~272 ms | |
#83 | Qwen3.5-27B | 27B | INT4 Q8 FP16 | 195 tok/s 125 tok/s 69 tok/s | ~265 ms ~268 ms ~274 ms | |
#84 | Qwen3.5-122B-A10B | 122B (10B active) | INT4 Q8 | 294 tok/s 226 tok/s | ~140 ms ~141 ms | |
#92 | Sarvam-30B | 32B (2.4B active) | INT4 Q8 FP16 | 502 tok/s 447 tok/s 357 tok/s | ~39 ms ~39 ms ~40 ms | |
#94 | Qwen3.6 35B A3B | 35B (3B active) | INT4 Q8 FP16 | 482 tok/s 419 tok/s 323 tok/s | ~45 ms ~45 ms ~46 ms | |
#96 | Muse Glimmer 30B | 30B | INT4 Q8 FP16 | 181 tok/s 114 tok/s 63 tok/s | ~293 ms ~296 ms ~304 ms | |
#100 | Phi-4 Reasoning Plus | 14B | INT4 Q8 FP16 | 238 tok/s 177 tok/s 112 tok/s | ~142 ms ~143 ms ~146 ms | |
#101 | MiMo V2 Flash | 15B (309B active) | INT4 Q8 FP16 | 24 tok/s 13 tok/s 7 tok/s | ~2.9s ~2.9s ~3.0s | |
#105 | Inkling-Small | 276B (12B active) | INT4 | 86 tok/s | ~208 ms | |
#106 | GLM-4.5-Air | 106B (12B active) | INT4 Q8 | 136 tok/s 121 tok/s | ~156 ms ~157 ms | |
#108 | Gemma 4 26B A4B | 25.2B (3.8B active) | INT4 Q8 FP16 | 465 tok/s 393 tok/s 291 tok/s | ~50 ms ~50 ms ~51 ms | |
#109 | Gemma 4 12B | 11.95B | INT4 Q8 FP16 | 260 tok/s 198 tok/s 127 tok/s | ~122 ms ~123 ms ~126 ms | |
#111 | MiniMax M2 | 229B (10B active) | INT4 | 108 tok/s | ~159 ms | |
#113 | Qwen3.5-9B | 9B | INT4 Q8 FP16 | 366 tok/s 270 tok/s 172 tok/s | ~92 ms ~93 ms ~95 ms | |
#114 | Qwen3.5-35B-A3B | 35B (3B active) | INT4 Q8 FP16 | 482 tok/s 419 tok/s 323 tok/s | ~45 ms ~45 ms ~46 ms | |
#114 | Ling 3.0 Flash | 124B (5.1B active) | INT4 Q8 | 441 tok/s 356 tok/s | ~94 ms ~95 ms | |
#118 | K2 Horizon 7B | 7B | INT4 Q8 FP16 | 406 tok/s 311 tok/s 206 tok/s | ~73 ms ~74 ms ~75 ms | |
#119 | Step 3.5 Flash | 196.81B (11B active) | INT4 Q8 | 277 tok/s 190 tok/s | ~155 ms ~182 ms | |
#126 | Qwen3 Next 80B A3B | 80B (79B active) | INT4 Q8 FP16 | 68 tok/s 43 tok/s 23 tok/s | ~763 ms ~772 ms ~857 ms | |
#128 | NVIDIA Nemotron 3 Nano 30B-A3B | 3.5B (30B active) | INT4 Q8 FP16 | 174 tok/s 112 tok/s 62 tok/s | ~284 ms ~287 ms ~295 ms | |
#129 | Gemma 4 31B | 30.7B | INT4 Q8 FP16 | 178 tok/s 112 tok/s 62 tok/s | ~300 ms ~303 ms ~310 ms | |
#130 | GPT-OSS 120B | 117B (5.1B active) | INT4 Q8 | 144 tok/s 140 tok/s | ~98 ms ~98 ms | |
#131 | Nemotron 3.5 Lightning | 30B (3B active) | INT4 Q8 FP16 | 300 tok/s 283 tok/s 236 tok/s | ~45 ms ~45 ms ~46 ms | |
#132 | Qwen3.5-4B | 4B | INT4 Q8 FP16 | 484 tok/s 401 tok/s 291 tok/s | ~44 ms ~45 ms ~46 ms | |
#133 | Qwen3-30B-A3B | 30B (3B active) | INT4 Q8 FP16 | 484 tok/s 421 tok/s 325 tok/s | ~44 ms ~44 ms ~45 ms | |
#136 | GPT-OSS 20B | 21B (3.6B active) | INT4 Q8 FP16 | 321 tok/s 293 tok/s 235 tok/s | ~48 ms ~48 ms ~49 ms | |
#139 | Qwen3-235B-A22B | 235B (22B active) | INT4 | 194 tok/s | ~266 ms | |
#144 | Mistral Medium 3.5 128B | 128B | INT4 Q8 | 44 tok/s 28 tok/s | ~1.2s ~1.2s | |
#146 | Granite 4.2 8B | 8B | INT4 Q8 FP16 | 385 tok/s 289 tok/s 187 tok/s | ~83 ms ~84 ms ~85 ms | |
#150 | Gemma 4 E4B | 8B | INT4 Q8 FP16 | 385 tok/s 289 tok/s 187 tok/s | ~83 ms ~84 ms ~85 ms | |
#151 | Nemotron 3.5 Lightning 30B A3B | 30B (3B active) | INT4 Q8 FP16 | 484 tok/s 421 tok/s 325 tok/s | ~44 ms ~44 ms ~45 ms | |
#153 | Qwen3-32B | 32B | INT4 Q8 FP16 | 173 tok/s 108 tok/s 60 tok/s | ~312 ms ~316 ms ~323 ms | |
#154 | Qwen2.5-72B | 72B | INT4 Q8 FP16 | 90 tok/s 53 tok/s 28 tok/s | ~693 ms ~701 ms ~718 ms | |
#155 | Mistral Large 3 | 41B (675B active) | INT4 Q8 FP16 | 11 tok/s 6 tok/s 3 tok/s | ~6.3s ~6.3s ~6.5s | |
#156 | GLM-4 | 32B | INT4 Q8 FP16 | 141 tok/s 96 tok/s 56 tok/s | ~313 ms ~317 ms ~324 ms | |
#157 | DeepSeek-R1 70B | 70B | INT4 Q8 FP16 | 95 tok/s 55 tok/s 29 tok/s | ~674 ms ~681 ms ~698 ms | |
#158 | GLM-4V | 9B | INT4 Q8 FP16 | 302 tok/s 237 tok/s 158 tok/s | ~93 ms ~94 ms ~96 ms | |
#159 | Gemma 3 27B | 27B | INT4 Q8 FP16 | 195 tok/s 125 tok/s 69 tok/s | ~265 ms ~268 ms ~274 ms | |
#160 | Phi-4 | 14B | INT4 Q8 FP16 | 293 tok/s 204 tok/s 122 tok/s | ~141 ms ~142 ms ~146 ms | |
#161 | Llama 3.3 70B | 70B | INT4 Q8 FP16 | 92 tok/s 55 tok/s 29 tok/s | ~674 ms ~682 ms ~698 ms | |
#162 | DeepSeek-R1 32B | 32B | INT4 Q8 FP16 | 177 tok/s 110 tok/s 60 tok/s | ~312 ms ~315 ms ~323 ms | |
#163 | Hunyuan Standard | 52B (389B active) | INT4 Q8 FP16 | 19 tok/s 11 tok/s 5 tok/s | ~3.6s ~3.7s ~3.7s | |
#164 | Qwen2.5-32B | 32B | INT4 Q8 FP16 | 173 tok/s 108 tok/s 60 tok/s | ~312 ms ~316 ms ~323 ms | |
#168 | Magistral Small | 24B | INT4 Q8 FP16 | 171 tok/s 120 tok/s 72 tok/s | ~237 ms ~240 ms ~246 ms | |
#169 | OLMo 3 7B Think | 7B | INT4 Q8 FP16 | 339 tok/s 273 tok/s 188 tok/s | ~73 ms ~74 ms ~76 ms | |
#170 | OLMo 3.1 32B Instruct | 32B | INT4 Q8 FP16 | 141 tok/s 96 tok/s 56 tok/s | ~313 ms ~317 ms ~324 ms | |
#171 | Llama 3.1 70B | 70B | INT4 Q8 FP16 | 92 tok/s 55 tok/s 29 tok/s | ~674 ms ~682 ms ~698 ms | |
#172 | DeepSeek-R1 14B | 14B | INT4 Q8 FP16 | 299 tok/s 206 tok/s 123 tok/s | ~141 ms ~142 ms ~145 ms | |
#173 | Ministral 3 14B | 14B | INT4 Q8 FP16 | 238 tok/s 177 tok/s 112 tok/s | ~142 ms ~143 ms ~146 ms | |
#174 | Mistral-Large-2407 | 123B | INT4 Q8 | 56 tok/s 32 tok/s | ~1.2s ~1.2s | |
#175 | Llama 4 Scout | 109B (17B active) | INT4 Q8 | 236 tok/s 166 tok/s | ~200 ms ~201 ms | |
#177 | Gemma 3 12B | 12B | INT4 Q8 FP16 | 319 tok/s 226 tok/s 138 tok/s | ~121 ms ~123 ms ~126 ms | |
#178 | Command A | 111B | INT4 Q8 | 49 tok/s 31 tok/s | ~1.1s ~1.1s | |
#179 | Qwen2-72B | 72B | INT4 Q8 FP16 | 90 tok/s 53 tok/s 28 tok/s | ~693 ms ~701 ms ~718 ms | |
#180 | OLMo 3 32B Base | 32B | INT4 Q8 FP16 | 141 tok/s 96 tok/s 56 tok/s | ~313 ms ~317 ms ~324 ms | |
#181 | Mistral-Small-2501 | 24B | INT4 Q8 FP16 | 211 tok/s 137 tok/s 77 tok/s | ~236 ms ~239 ms ~245 ms | |
#182 | Ministral 3 8B | 8B | INT4 Q8 FP16 | 321 tok/s 254 tok/s 172 tok/s | ~83 ms ~84 ms ~86 ms | |
#183 | Qwen3.5-2B | 2B | INT4 Q8 FP16 | 556 tok/s 497 tok/s 403 tok/s | ~25 ms ~25 ms ~26 ms | |
#184 | MiniCPM5-2B | 2B | INT4 Q8 FP16 | 556 tok/s 497 tok/s 403 tok/s | ~25 ms ~25 ms ~26 ms | |
#185 | Qwen2.5-14B | 14B | INT4 Q8 FP16 | 293 tok/s 204 tok/s 122 tok/s | ~141 ms ~142 ms ~146 ms | |
#186 | Gemma 2 27B | 27B | INT4 Q8 FP16 | 195 tok/s 125 tok/s 69 tok/s | ~265 ms ~268 ms ~274 ms | |
#187 | OLMo 3.1 32B Think | 32B | INT4 Q8 FP16 | 141 tok/s 96 tok/s 56 tok/s | ~313 ms ~317 ms ~324 ms | |
#188 | Gemma 4 E2B | 5.1B | INT4 Q8 FP16 | 452 tok/s 363 tok/s 253 tok/s | ~55 ms ~55 ms ~57 ms | |
#189 | Qwen3-4B | 4B | INT4 Q8 FP16 | 484 tok/s 401 tok/s 291 tok/s | ~44 ms ~45 ms ~46 ms | |
#190 | DeepSeek-R1 8B | 8B | INT4 Q8 FP16 | 390 tok/s 292 tok/s 188 tok/s | ~83 ms ~83 ms ~85 ms | |
#191 | Llama 3 70B | 70B | INT4 Q8 FP16 | 92 tok/s 55 tok/s 29 tok/s | ~674 ms ~682 ms ~698 ms | |
#192 | Command R Plus | 104B | INT4 Q8 | 53 tok/s 33 tok/s | ~1.0s ~1.0s | |
#193 | Ministral 3 3B | 3B | INT4 Q8 FP16 | 442 tok/s 391 tok/s 306 tok/s | ~35 ms ~35 ms ~36 ms | |
#194 | ERNIE-4.5-300B-A47B | 300B (47B active) | INT4 | 110 tok/s | ~560 ms | |
#195 | Qwen3-14B | 14B | INT4 Q8 FP16 | 293 tok/s 204 tok/s 122 tok/s | ~141 ms ~142 ms ~146 ms | |
#196 | Kimi Linear 48B A3B Instruct | 48B (3B active) | INT4 Q8 FP16 | 271 tok/s 258 tok/s 218 tok/s | ~49 ms ~49 ms ~50 ms | |
#197 | Gemma 2 9B | 9B | INT4 Q8 FP16 | 366 tok/s 270 tok/s 172 tok/s | ~92 ms ~93 ms ~95 ms | |
#198 | Qwen2.5-7B | 7B | INT4 Q8 FP16 | 406 tok/s 311 tok/s 206 tok/s | ~73 ms ~74 ms ~75 ms | |
#199 | Qwen3-8B | 8B | INT4 Q8 FP16 | 385 tok/s 289 tok/s 187 tok/s | ~83 ms ~84 ms ~85 ms | |
#200 | Gemma 3 4B | 4B | INT4 Q8 FP16 | 484 tok/s 401 tok/s 291 tok/s | ~44 ms ~45 ms ~46 ms | |
#201 | DeepSeek-R1 7B | 7B | INT4 Q8 FP16 | 411 tok/s 314 tok/s 207 tok/s | ~73 ms ~74 ms ~75 ms | |
#202 | Llama 3.1 8B | 8B | INT4 Q8 FP16 | 385 tok/s 289 tok/s 187 tok/s | ~83 ms ~84 ms ~85 ms | |
#203 | Mixtral-8x22B-v0.1 | 176B (22B active) | INT4 Q8 | 202 tok/s 131 tok/s | ~251 ms ~274 ms | |
#204 | Qwen3-1.7B | 1.7B | INT4 Q8 FP16 | 570 tok/s 517 tok/s 428 tok/s | ~22 ms ~22 ms ~22 ms | |
#205 | Command R | 35B | INT4 Q8 FP16 | 133 tok/s 90 tok/s 51 tok/s | ~342 ms ~345 ms ~353 ms | |
#206 | Gemma 3 270M | 0.27B | INT4 Q8 FP16 | 637 tok/s 627 tok/s 603 tok/s | ~6 ms ~6 ms ~6 ms | |
#208 | Llama 3 8B | 8B | INT4 Q8 FP16 | 385 tok/s 289 tok/s 187 tok/s | ~83 ms ~84 ms ~85 ms | |
#209 | Qwen3-0.6B | 0.6B | INT4 Q8 FP16 | 629 tok/s 605 tok/s 557 tok/s | ~10 ms ~10 ms ~10 ms | |
#210 | Phi-4-Mini | 3.8B | INT4 Q8 FP16 | 491 tok/s 409 tok/s 299 tok/s | ~42 ms ~43 ms ~44 ms | |
#211 | Mixtral-8x7B-v0.1 | 46.7B (7B active) | INT4 Q8 FP16 | 381 tok/s 298 tok/s 200 tok/s | ~84 ms ~85 ms ~87 ms | |
#212 | Ministral-8B-2410 | 8B | INT4 Q8 FP16 | 385 tok/s 289 tok/s 187 tok/s | ~83 ms ~84 ms ~85 ms | |
#213 | DeepSeek-R1 1.5B | 1.5B | INT4 Q8 FP16 | 579 tok/s 531 tok/s 447 tok/s | ~20 ms ~20 ms ~20 ms | |
#214 | Llama 3.2 3B | 3B | INT4 Q8 FP16 | 518 tok/s 444 tok/s 338 tok/s | ~35 ms ~35 ms ~36 ms | |
#215 | Qwen3.5-0.8B | 0.8B | INT4 Q8 FP16 | 617 tok/s 586 tok/s 528 tok/s | ~12 ms ~12 ms ~12 ms | |
#216 | Phi-3-mini | 3.8B | INT4 Q8 FP16 | 491 tok/s 409 tok/s 299 tok/s | ~42 ms ~43 ms ~44 ms | |
#217 | Gemma 3n E2B IT | 6B (2B active) | INT4 Q8 FP16 | 442 tok/s 410 tok/s 344 tok/s | ~27 ms ~27 ms ~28 ms | |
#218 | Mistral-7B-Instruct-v0.1 | 7.3B | INT4 Q8 FP16 | 399 tok/s 304 tok/s 200 tok/s | ~76 ms ~77 ms ~78 ms | |
#219 | Llama 3.2 1B | 1B | INT4 Q8 FP16 | 605 tok/s 568 tok/s 501 tok/s | ~15 ms ~15 ms ~15 ms | |
#220 | Gemma 2 2B | 2B | INT4 Q8 FP16 | 556 tok/s 497 tok/s 403 tok/s | ~25 ms ~25 ms ~26 ms | |
#221 | Phi-3-medium | 14B | INT4 Q8 FP16 | 293 tok/s 204 tok/s 122 tok/s | ~141 ms ~142 ms ~146 ms | |
#222 | OLMo 3 7B Instruct | 7B | INT4 Q8 FP16 | 339 tok/s 273 tok/s 188 tok/s | ~73 ms ~74 ms ~76 ms | |
#223 | Yi-34B | 34B | INT4 Q8 FP16 | 135 tok/s 92 tok/s 53 tok/s | ~332 ms ~336 ms ~344 ms | |
#224 | Qwen2-7B | 7B | INT4 Q8 FP16 | 406 tok/s 311 tok/s 206 tok/s | ~73 ms ~74 ms ~75 ms | |
#225 | Phi-3-small | 7B | INT4 Q8 FP16 | 406 tok/s 311 tok/s 206 tok/s | ~73 ms ~74 ms ~75 ms | |
#226 | Gemma 3 1B | 1B | INT4 Q8 FP16 | 605 tok/s 568 tok/s 501 tok/s | ~15 ms ~15 ms ~15 ms | |
#227 | Mistral-7B-Instruct-v0.2 | 7.3B | INT4 Q8 FP16 | 399 tok/s 304 tok/s 200 tok/s | ~76 ms ~77 ms ~78 ms | |
#228 | Falcon-180B | 180B | INT4 Q8 | 40 tok/s 22 tok/s | ~1.7s ~1.9s | |
#229 | Gemma 1 7B | 7B | INT4 Q8 FP16 | 339 tok/s 273 tok/s 188 tok/s | ~73 ms ~74 ms ~76 ms | |
#230 | Gemma 1 2B | 2B | INT4 Q8 FP16 | 566 tok/s 505 tok/s 408 tok/s | ~25 ms ~25 ms ~26 ms | |
#231 | ChatGLM3-6B | 6B | INT4 Q8 FP16 | 360 tok/s 296 tok/s 208 tok/s | ~64 ms ~64 ms ~66 ms | |
#232 | ChatGLM2-6B | 6B | INT4 Q8 FP16 | 360 tok/s 296 tok/s 208 tok/s | ~64 ms ~64 ms ~66 ms | |
#233 | ChatGLM-6B | 6B | INT4 Q8 FP16 | 360 tok/s 296 tok/s 208 tok/s | ~64 ms ~64 ms ~66 ms | |
Unranked Models (Sorted by Speed) | ||||||
- | Qwen2-0.5B | 0.5B | INT4 Q8 FP16 | 636 tok/s 615 tok/s 573 tok/s | ~9 ms ~9 ms ~9 ms | |
- | Qwen2.5-0.5B | 0.5B | INT4 Q8 FP16 | 636 tok/s 615 tok/s 573 tok/s | ~9 ms ~9 ms ~9 ms | |
- | ERNIE-4.5-0.3B | 0.3B | INT4 Q8 FP16 | 633 tok/s 622 tok/s 596 tok/s | ~7 ms ~7 ms ~7 ms | |
- | ERNIE-4.5-0.3B-Base | 0.3B | INT4 Q8 FP16 | 633 tok/s 622 tok/s 596 tok/s | ~7 ms ~7 ms ~7 ms | |
- | Falcon-1B | 1B | INT4 Q8 FP16 | 612 tok/s 574 tok/s 506 tok/s | ~15 ms ~15 ms ~15 ms | |
- | Falcon3-1B | 1B | INT4 Q8 FP16 | 605 tok/s 568 tok/s 501 tok/s | ~15 ms ~15 ms ~15 ms | |
- | Qwen2-1.5B | 1.5B | INT4 Q8 FP16 | 579 tok/s 531 tok/s 447 tok/s | ~20 ms ~20 ms ~20 ms | |
- | Qwen2.5-1.5B | 1.5B | INT4 Q8 FP16 | 579 tok/s 531 tok/s 447 tok/s | ~20 ms ~20 ms ~20 ms | |
- | Spark X2.5 1.7B | 1.7B | INT4 Q8 FP16 | 570 tok/s 517 tok/s 428 tok/s | ~22 ms ~22 ms ~22 ms | |
- | Hy-MT2 1.8B | 1.8B | INT4 Q8 FP16 | 565 tok/s 510 tok/s 419 tok/s | ~23 ms ~23 ms ~24 ms | |
- | Falcon-3B | 3B | INT4 Q8 FP16 | 527 tok/s 451 tok/s 342 tok/s | ~35 ms ~35 ms ~36 ms | |
- | CroissantLLM Base | 1.3B | INT4 Q8 FP16 | 524 tok/s 493 tok/s 428 tok/s | ~18 ms ~18 ms ~18 ms | |
- | Phi-1 | 1.3B | INT4 Q8 FP16 | 524 tok/s 493 tok/s 428 tok/s | ~18 ms ~18 ms ~18 ms | |
- | Phi-1.5 | 1.3B | INT4 Q8 FP16 | 524 tok/s 493 tok/s 428 tok/s | ~18 ms ~18 ms ~18 ms | |
- | DeepSeek-R1 3B | 3B | INT4 Q8 FP16 | 520 tok/s 446 tok/s 339 tok/s | ~35 ms ~35 ms ~36 ms | |
- | Falcon3-3B | 3B | INT4 Q8 FP16 | 518 tok/s 444 tok/s 338 tok/s | ~35 ms ~35 ms ~36 ms | |
- | Qwen2.5-3B | 3B | INT4 Q8 FP16 | 518 tok/s 444 tok/s 338 tok/s | ~35 ms ~35 ms ~36 ms | |
- | ERNIE-4.5-21B-A3B | 21B (3B active) | INT4 Q8 FP16 | 493 tok/s 427 tok/s 328 tok/s | ~41 ms ~42 ms ~42 ms | |
- | ERNIE-4.5-21B-A3B-Base | 21B (3B active) | INT4 Q8 FP16 | 493 tok/s 427 tok/s 328 tok/s | ~41 ms ~42 ms ~42 ms | |
- | ERNIE-4.5-VL-28B-A3B | 28B (3B active) | INT4 Q8 FP16 | 486 tok/s 423 tok/s 325 tok/s | ~43 ms ~44 ms ~44 ms | |
- | ERNIE-4.5-VL-28B-A3B-Base | 28B (3B active) | INT4 Q8 FP16 | 486 tok/s 423 tok/s 325 tok/s | ~43 ms ~44 ms ~44 ms | |
- | Hy-MT2 30B A3B | 30B (3B active) | INT4 Q8 FP16 | 484 tok/s 421 tok/s 325 tok/s | ~44 ms ~44 ms ~45 ms | |
- | Spark X2.5 4B | 4B | INT4 Q8 FP16 | 484 tok/s 401 tok/s 291 tok/s | ~44 ms ~45 ms ~46 ms | |
- | Phi-2 | 2.7B | INT4 Q8 FP16 | 452 tok/s 404 tok/s 321 tok/s | ~32 ms ~32 ms ~33 ms | |
- | MaLLaM-3B | 3B | INT4 Q8 FP16 | 442 tok/s 391 tok/s 306 tok/s | ~35 ms ~35 ms ~36 ms | |
- | SmolLM3 3B | 3B | INT4 Q8 FP16 | 442 tok/s 391 tok/s 306 tok/s | ~35 ms ~35 ms ~36 ms | |
- | AliceAI Foundation 80B A3B Base | 80B (3B active) | INT4 Q8 FP16 | 439 tok/s 389 tok/s 290 tok/s | ~59 ms ~59 ms ~65 ms | |
- | Falcon-7B | 7B | INT4 Q8 FP16 | 414 tok/s 316 tok/s 208 tok/s | ~73 ms ~74 ms ~75 ms | |
- | Falcon3-7B | 7B | INT4 Q8 FP16 | 406 tok/s 311 tok/s 206 tok/s | ~73 ms ~74 ms ~75 ms | |
- | Hy-MT2 7B | 7B | INT4 Q8 FP16 | 406 tok/s 311 tok/s 206 tok/s | ~73 ms ~74 ms ~75 ms | |
- | Mistral-7B-v0.1 | 7.3B | INT4 Q8 FP16 | 399 tok/s 304 tok/s 200 tok/s | ~76 ms ~77 ms ~78 ms | |
- | LensVLM 9B | 9B | INT4 Q8 FP16 | 366 tok/s 270 tok/s 172 tok/s | ~92 ms ~93 ms ~95 ms | |
- | MiMo V2.6 Distill Qwen 9B | 9B | INT4 Q8 FP16 | 366 tok/s 270 tok/s 172 tok/s | ~92 ms ~93 ms ~95 ms | |
- | ChatGLM3-6B-32K | 6B | INT4 Q8 FP16 | 360 tok/s 296 tok/s 208 tok/s | ~64 ms ~64 ms ~66 ms | |
- | Yi-6B | 6B | INT4 Q8 FP16 | 360 tok/s 296 tok/s 208 tok/s | ~64 ms ~64 ms ~66 ms | |
- | Falcon3-10B | 10B | INT4 Q8 FP16 | 349 tok/s 254 tok/s 159 tok/s | ~102 ms ~103 ms ~105 ms | |
- | Kimi-VL-A3B-Instruct | 16B (3B active) | INT4 Q8 FP16 | 349 tok/s 322 tok/s 262 tok/s | ~41 ms ~41 ms ~42 ms | |
- | Falcon2-11B | 11B | INT4 Q8 FP16 | 341 tok/s 243 tok/s 149 tok/s | ~112 ms ~113 ms ~115 ms | |
- | Hunyuan Lite | 7B | INT4 Q8 FP16 | 339 tok/s 273 tok/s 188 tok/s | ~73 ms ~74 ms ~76 ms | |
- | OLMo 3 7B Base | 7B | INT4 Q8 FP16 | 339 tok/s 273 tok/s 188 tok/s | ~73 ms ~74 ms ~76 ms | |
- | SEA-LION-7B | 7.1B | INT4 Q8 FP16 | 338 tok/s 272 tok/s 187 tok/s | ~74 ms ~75 ms ~77 ms | |
- | SEA-LION-7B-Instruct | 7.1B | INT4 Q8 FP16 | 338 tok/s 272 tok/s 187 tok/s | ~74 ms ~75 ms ~77 ms | |
- | Sahabat-AI-Llama3-8B-Instruct | 8B | INT4 Q8 FP16 | 321 tok/s 254 tok/s 172 tok/s | ~83 ms ~84 ms ~86 ms | |
- | Typhoon-2-8B | 8B | INT4 Q8 FP16 | 321 tok/s 254 tok/s 172 tok/s | ~83 ms ~84 ms ~86 ms | |
- | Laguna S 2.1 | 118B (8B active) | INT4 Q8 | 319 tok/s 254 tok/s | ~120 ms ~121 ms | |
- | GLM-4-9B | 9B | INT4 Q8 FP16 | 302 tok/s 237 tok/s 158 tok/s | ~93 ms ~94 ms ~96 ms | |
- | GLM-4-9B-Chat | 9B | INT4 Q8 FP16 | 302 tok/s 237 tok/s 158 tok/s | ~93 ms ~94 ms ~96 ms | |
- | GLM-4-9B-Chat-1M | 9B | INT4 Q8 FP16 | 302 tok/s 237 tok/s 158 tok/s | ~93 ms ~94 ms ~96 ms | |
- | Yi-9B | 9B | INT4 Q8 FP16 | 302 tok/s 237 tok/s 158 tok/s | ~93 ms ~94 ms ~96 ms | |
- | Sahabat-AI-Gemma2-9B | 9.2B | INT4 Q8 FP16 | 300 tok/s 234 tok/s 155 tok/s | ~95 ms ~96 ms ~98 ms | |
- | Sahabat-AI-Gemma2-9B-Instruct | 9.2B | INT4 Q8 FP16 | 300 tok/s 234 tok/s 155 tok/s | ~95 ms ~96 ms ~98 ms | |
- | GreenMind-14B-R1 | 14B | INT4 Q8 FP16 | 238 tok/s 177 tok/s 112 tok/s | ~142 ms ~143 ms ~146 ms | |
- | Hunyuan A13B | 80B (13B active) | INT4 Q8 FP16 | 159 tok/s 134 tok/s 90 tok/s | ~154 ms ~155 ms ~172 ms | |
- | Falcon-40B | 40B | INT4 Q8 FP16 | 150 tok/s 91 tok/s 49 tok/s | ~387 ms ~392 ms ~401 ms | |
- | ERNIE-4.5-300B-A47B-Base | 300B (47B active) | INT4 | 110 tok/s | ~560 ms | |
- | Typhoon-2-70B | 70B | INT4 Q8 FP16 | 76 tok/s 49 tok/s 27 tok/s | ~677 ms ~684 ms ~700 ms | |
- | Kimi-Dev-72B | 72B | INT4 Q8 FP16 | 74 tok/s 47 tok/s 25 tok/s | ~696 ms ~703 ms ~781 ms | |
- | MiMo V2.6 Flash RL | 309B (15B active) | INT4 | 73 tok/s | ~269 ms | |
- | GLM-130B | 130B | INT4 Q8 | 43 tok/s 27 tok/s | ~1.3s ~1.3s | |
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 Instinct MI300X (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 5300 GB/s aggregate bandwidth, batch size 1 inference operates in a memory-bandwidth bound regime. Generates approximately 1178 tok/s on an 8B Q4 model and 139 tok/s on a 70B Q4 model.
Frequently Asked Questions
©2025 ApX Machine Learning