Instinct MI250 (128GB)
The Instinct MI250 (128GB) provides 128 GB of memory with 3276 GB/s bandwidth, supporting local inference via ROCm and Vulkan acceleration.
Memory
128 GB
Usable Memory Ceiling
128 GB
Bandwidth
3276 GB/s
Max Inference Class
70B+ Models (Q4)
TDP
500W
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 MI250 (128GB)
Inference throughput, memory compatibility, and generation speeds for open-weights LLMs on Instinct MI250 (128GB).
Sort by:
Context:
KV Cache:
GPUs:
| Rank | Model | Parameters | Precision | TPS (Approx.) | TTFT | Can Run |
|---|---|---|---|---|---|---|
#33 | Qwen 3.8 27B | 27B | INT4 Q8 FP16 | 136 tok/s 83 tok/s 45 tok/s | ~928 ms ~932 ms ~943 ms | |
#41 | Qwen 3.8 Flash | 176B (6B active) | INT4 | 360 tok/s | ~202 ms | |
#48 | Agnes 3.0 Flash | 33B | INT4 Q8 FP16 | 116 tok/s 70 tok/s 37 tok/s | ~1.1s ~1.1s ~1.1s | |
#61 | Sarvam-105B | 106B (10.3B active) | INT4 Q8 | 262 tok/s 170 tok/s | ~476 ms ~518 ms | |
#72 | K2 Horizon MoVA 36B A4B | 36B (4B active) | INT4 Q8 FP16 | 378 tok/s 301 tok/s 206 tok/s | ~184 ms ~185 ms ~186 ms | |
#75 | Ling 3.0 Flash VL | 124B (5.5B active) | INT4 Q8 | 354 tok/s 239 tok/s | ~339 ms ~400 ms | |
#83 | Qwen3.5-27B | 27B | INT4 Q8 FP16 | 136 tok/s 83 tok/s 45 tok/s | ~928 ms ~932 ms ~943 ms | |
#84 | Qwen3.5-122B-A10B | 122B (10B active) | INT4 Q8 | 219 tok/s 144 tok/s | ~487 ms ~574 ms | |
#92 | Sarvam-30B | 32B (2.4B active) | INT4 Q8 FP16 | 435 tok/s 371 tok/s 277 tok/s | ~129 ms ~129 ms ~130 ms | |
#94 | Qwen3.6 35B A3B | 35B (3B active) | INT4 Q8 FP16 | 411 tok/s 341 tok/s 245 tok/s | ~150 ms ~151 ms ~152 ms | |
#96 | Muse Glimmer 30B | 30B | INT4 Q8 FP16 | 125 tok/s 76 tok/s 41 tok/s | ~1.0s ~1.0s ~1.0s | |
#100 | Phi-4 Reasoning Plus | 14B | INT4 Q8 FP16 | 170 tok/s 122 tok/s 74 tok/s | ~490 ms ~493 ms ~498 ms | |
#101 | MiMo V2 Flash | 15B (309B active) | INT4 Q8 FP16 | 15 tok/s 8 tok/s 4 tok/s | ~10.1s ~10.2s ~10.3s | |
#106 | GLM-4.5-Air | 106B (12B active) | INT4 Q8 | 91 tok/s 72 tok/s | ~538 ms ~633 ms | |
#108 | Gemma 4 26B A4B | 25.2B (3.8B active) | INT4 Q8 FP16 | 392 tok/s 313 tok/s 216 tok/s | ~168 ms ~169 ms ~170 ms | |
#109 | Gemma 4 12B | 11.95B | INT4 Q8 FP16 | 189 tok/s 138 tok/s 85 tok/s | ~420 ms ~422 ms ~427 ms | |
#113 | Qwen3.5-9B | 9B | INT4 Q8 FP16 | 286 tok/s 198 tok/s 118 tok/s | ~318 ms ~320 ms ~323 ms | |
#114 | Qwen3.5-35B-A3B | 35B (3B active) | INT4 Q8 FP16 | 411 tok/s 341 tok/s 245 tok/s | ~150 ms ~151 ms ~152 ms | |
#114 | Ling 3.0 Flash | 124B (5.1B active) | INT4 Q8 | 364 tok/s 249 tok/s | ~326 ms ~384 ms | |
#118 | K2 Horizon 7B | 7B | INT4 Q8 FP16 | 326 tok/s 234 tok/s 144 tok/s | ~249 ms ~251 ms ~253 ms | |
#119 | Step 3.5 Flash | 196.81B (11B active) | INT4 | 194 tok/s | ~586 ms | |
#126 | Qwen3 Next 80B A3B | 80B (79B active) | INT4 Q8 | 44 tok/s 28 tok/s | ~2.7s ~2.7s | |
#128 | NVIDIA Nemotron 3 Nano 30B-A3B | 3.5B (30B active) | INT4 Q8 FP16 | 119 tok/s 74 tok/s 40 tok/s | ~996 ms ~1.0s ~1.0s | |
#129 | Gemma 4 31B | 30.7B | INT4 Q8 FP16 | 123 tok/s 74 tok/s 40 tok/s | ~1.1s ~1.1s ~1.1s | |
#130 | GPT-OSS 120B | 117B (5.1B active) | INT4 | 97 tok/s | ~329 ms | |
#131 | Nemotron 3.5 Lightning | 30B (3B active) | INT4 Q8 FP16 | 224 tok/s 209 tok/s 168 tok/s | ~149 ms ~150 ms ~151 ms | |
#132 | Qwen3.5-4B | 4B | INT4 Q8 FP16 | 414 tok/s 322 tok/s 216 tok/s | ~147 ms ~148 ms ~150 ms | |
#133 | Qwen3-30B-A3B | 30B (3B active) | INT4 Q8 FP16 | 414 tok/s 343 tok/s 246 tok/s | ~147 ms ~148 ms ~149 ms | |
#136 | GPT-OSS 20B | 21B (3.6B active) | INT4 Q8 FP16 | 243 tok/s 218 tok/s 168 tok/s | ~159 ms ~160 ms ~161 ms | |
#139 | Qwen3-235B-A22B | 235B (22B active) | INT4 | 122 tok/s | ~1.1s | |
#144 | Mistral Medium 3.5 128B | 128B | INT4 | 28 tok/s | ~4.3s | |
#146 | Granite 4.2 8B | 8B | INT4 Q8 FP16 | 305 tok/s 214 tok/s 130 tok/s | ~283 ms ~285 ms ~288 ms | |
#150 | Gemma 4 E4B | 8B | INT4 Q8 FP16 | 305 tok/s 214 tok/s 130 tok/s | ~283 ms ~285 ms ~288 ms | |
#151 | Nemotron 3.5 Lightning 30B A3B | 30B (3B active) | INT4 Q8 FP16 | 414 tok/s 343 tok/s 246 tok/s | ~147 ms ~148 ms ~149 ms | |
#153 | Qwen3-32B | 32B | INT4 Q8 FP16 | 119 tok/s 71 tok/s 38 tok/s | ~1.1s ~1.1s ~1.1s | |
#154 | Qwen2.5-72B | 72B | INT4 Q8 | 59 tok/s 34 tok/s | ~2.4s ~2.5s | |
#155 | Mistral Large 3 | 41B (675B active) | INT4 Q8 FP16 | 7 tok/s 4 tok/s 2 tok/s | ~22.1s ~22.2s ~22.5s | |
#156 | GLM-4 | 32B | INT4 Q8 FP16 | 95 tok/s 63 tok/s 36 tok/s | ~1.1s ~1.1s ~1.1s | |
#157 | DeepSeek-R1 70B | 70B | INT4 Q8 | 62 tok/s 35 tok/s | ~2.4s ~2.4s | |
#158 | GLM-4V | 9B | INT4 Q8 FP16 | 226 tok/s 169 tok/s 107 tok/s | ~319 ms ~320 ms ~324 ms | |
#159 | Gemma 3 27B | 27B | INT4 Q8 FP16 | 136 tok/s 83 tok/s 45 tok/s | ~928 ms ~932 ms ~943 ms | |
#160 | Phi-4 | 14B | INT4 Q8 FP16 | 218 tok/s 143 tok/s 81 tok/s | ~489 ms ~492 ms ~497 ms | |
#161 | Llama 3.3 70B | 70B | INT4 Q8 | 60 tok/s 35 tok/s | ~2.4s ~2.4s | |
#162 | DeepSeek-R1 32B | 32B | INT4 Q8 FP16 | 121 tok/s 72 tok/s 38 tok/s | ~1.1s ~1.1s ~1.1s | |
#163 | Hunyuan Standard | 52B (389B active) | INT4 Q8 FP16 | 12 tok/s 7 tok/s 3 tok/s | ~12.8s ~12.8s ~14.1s | |
#164 | Qwen2.5-32B | 32B | INT4 Q8 FP16 | 119 tok/s 71 tok/s 38 tok/s | ~1.1s ~1.1s ~1.1s | |
#168 | Magistral Small | 24B | INT4 Q8 FP16 | 117 tok/s 80 tok/s 46 tok/s | ~829 ms ~833 ms ~842 ms | |
#169 | OLMo 3 7B Think | 7B | INT4 Q8 FP16 | 260 tok/s 200 tok/s 131 tok/s | ~250 ms ~251 ms ~254 ms | |
#170 | OLMo 3.1 32B Instruct | 32B | INT4 Q8 FP16 | 95 tok/s 63 tok/s 36 tok/s | ~1.1s ~1.1s ~1.1s | |
#171 | Llama 3.1 70B | 70B | INT4 Q8 | 60 tok/s 35 tok/s | ~2.4s ~2.4s | |
#172 | DeepSeek-R1 14B | 14B | INT4 Q8 FP16 | 223 tok/s 145 tok/s 82 tok/s | ~489 ms ~491 ms ~497 ms | |
#173 | Ministral 3 14B | 14B | INT4 Q8 FP16 | 170 tok/s 122 tok/s 74 tok/s | ~490 ms ~493 ms ~498 ms | |
#174 | Mistral-Large-2407 | 123B | INT4 Q8 | 36 tok/s 18 tok/s | ~4.2s ~4.9s | |
#175 | Llama 4 Scout | 109B (17B active) | INT4 Q8 | 169 tok/s 108 tok/s | ~698 ms ~761 ms | |
#177 | Gemma 3 12B | 12B | INT4 Q8 FP16 | 241 tok/s 161 tok/s 93 tok/s | ~421 ms ~423 ms ~427 ms | |
#178 | Command A | 111B | INT4 Q8 | 31 tok/s 18 tok/s | ~3.8s ~4.5s | |
#179 | Qwen2-72B | 72B | INT4 Q8 | 59 tok/s 34 tok/s | ~2.4s ~2.5s | |
#180 | OLMo 3 32B Base | 32B | INT4 Q8 FP16 | 95 tok/s 63 tok/s 36 tok/s | ~1.1s ~1.1s ~1.1s | |
#181 | Mistral-Small-2501 | 24B | INT4 Q8 FP16 | 149 tok/s 92 tok/s 50 tok/s | ~827 ms ~831 ms ~840 ms | |
#182 | Ministral 3 8B | 8B | INT4 Q8 FP16 | 243 tok/s 184 tok/s 118 tok/s | ~284 ms ~285 ms ~289 ms | |
#183 | Qwen3.5-2B | 2B | INT4 Q8 FP16 | 504 tok/s 430 tok/s 324 tok/s | ~80 ms ~80 ms ~81 ms | |
#184 | MiniCPM5-2B | 2B | INT4 Q8 FP16 | 504 tok/s 430 tok/s 324 tok/s | ~80 ms ~80 ms ~81 ms | |
#185 | Qwen2.5-14B | 14B | INT4 Q8 FP16 | 218 tok/s 143 tok/s 81 tok/s | ~489 ms ~492 ms ~497 ms | |
#186 | Gemma 2 27B | 27B | INT4 Q8 FP16 | 136 tok/s 83 tok/s 45 tok/s | ~928 ms ~932 ms ~943 ms | |
#187 | OLMo 3.1 32B Think | 32B | INT4 Q8 FP16 | 95 tok/s 63 tok/s 36 tok/s | ~1.1s ~1.1s ~1.1s | |
#188 | Gemma 4 E2B | 5.1B | INT4 Q8 FP16 | 377 tok/s 283 tok/s 183 tok/s | ~185 ms ~186 ms ~188 ms | |
#189 | Qwen3-4B | 4B | INT4 Q8 FP16 | 414 tok/s 322 tok/s 216 tok/s | ~147 ms ~148 ms ~150 ms | |
#190 | DeepSeek-R1 8B | 8B | INT4 Q8 FP16 | 310 tok/s 217 tok/s 130 tok/s | ~283 ms ~285 ms ~288 ms | |
#191 | Llama 3 70B | 70B | INT4 Q8 | 60 tok/s 35 tok/s | ~2.4s ~2.4s | |
#192 | Command R Plus | 104B | INT4 Q8 | 33 tok/s 20 tok/s | ~3.5s ~3.9s | |
#193 | Ministral 3 3B | 3B | INT4 Q8 FP16 | 366 tok/s 312 tok/s 230 tok/s | ~114 ms ~114 ms ~116 ms | |
#195 | Qwen3-14B | 14B | INT4 Q8 FP16 | 218 tok/s 143 tok/s 81 tok/s | ~489 ms ~492 ms ~497 ms | |
#196 | Kimi Linear 48B A3B Instruct | 48B (3B active) | INT4 Q8 FP16 | 198 tok/s 187 tok/s 147 tok/s | ~161 ms ~162 ms ~177 ms | |
#197 | Gemma 2 9B | 9B | INT4 Q8 FP16 | 286 tok/s 198 tok/s 118 tok/s | ~318 ms ~320 ms ~323 ms | |
#198 | Qwen2.5-7B | 7B | INT4 Q8 FP16 | 326 tok/s 234 tok/s 144 tok/s | ~249 ms ~251 ms ~253 ms | |
#199 | Qwen3-8B | 8B | INT4 Q8 FP16 | 305 tok/s 214 tok/s 130 tok/s | ~283 ms ~285 ms ~288 ms | |
#200 | Gemma 3 4B | 4B | INT4 Q8 FP16 | 414 tok/s 322 tok/s 216 tok/s | ~147 ms ~148 ms ~150 ms | |
#201 | DeepSeek-R1 7B | 7B | INT4 Q8 FP16 | 332 tok/s 236 tok/s 145 tok/s | ~249 ms ~251 ms ~253 ms | |
#202 | Llama 3.1 8B | 8B | INT4 Q8 FP16 | 305 tok/s 214 tok/s 130 tok/s | ~283 ms ~285 ms ~288 ms | |
#203 | Mixtral-8x22B-v0.1 | 176B (22B active) | INT4 | 142 tok/s | ~879 ms | |
#204 | Qwen3-1.7B | 1.7B | INT4 Q8 FP16 | 523 tok/s 454 tok/s 350 tok/s | ~69 ms ~69 ms ~70 ms | |
#205 | Command R | 35B | INT4 Q8 FP16 | 89 tok/s 58 tok/s 33 tok/s | ~1.2s ~1.2s ~1.2s | |
#206 | Gemma 3 270M | 0.27B | INT4 Q8 FP16 | 620 tok/s 604 tok/s 569 tok/s | ~13 ms ~14 ms ~14 ms | |
#208 | Llama 3 8B | 8B | INT4 Q8 FP16 | 305 tok/s 214 tok/s 130 tok/s | ~283 ms ~285 ms ~288 ms | |
#209 | Qwen3-0.6B | 0.6B | INT4 Q8 FP16 | 608 tok/s 572 tok/s 506 tok/s | ~26 ms ~26 ms ~26 ms | |
#210 | Phi-4-Mini | 3.8B | INT4 Q8 FP16 | 422 tok/s 330 tok/s 223 tok/s | ~141 ms ~141 ms ~143 ms | |
#211 | Mixtral-8x7B-v0.1 | 46.7B (7B active) | INT4 Q8 FP16 | 302 tok/s 222 tok/s 139 tok/s | ~289 ms ~290 ms ~293 ms | |
#212 | Ministral-8B-2410 | 8B | INT4 Q8 FP16 | 305 tok/s 214 tok/s 130 tok/s | ~283 ms ~285 ms ~288 ms | |
#213 | DeepSeek-R1 1.5B | 1.5B | INT4 Q8 FP16 | 536 tok/s 471 tok/s 371 tok/s | ~61 ms ~61 ms ~62 ms | |
#214 | Llama 3.2 3B | 3B | INT4 Q8 FP16 | 455 tok/s 368 tok/s 259 tok/s | ~113 ms ~114 ms ~115 ms | |
#215 | Qwen3.5-0.8B | 0.8B | INT4 Q8 FP16 | 590 tok/s 546 tok/s 468 tok/s | ~34 ms ~34 ms ~35 ms | |
#216 | Phi-3-mini | 3.8B | INT4 Q8 FP16 | 422 tok/s 330 tok/s 223 tok/s | ~141 ms ~141 ms ~143 ms | |
#217 | Gemma 3n E2B IT | 6B (2B active) | INT4 Q8 FP16 | 366 tok/s 331 tok/s 264 tok/s | ~86 ms ~86 ms ~87 ms | |
#218 | Mistral-7B-Instruct-v0.1 | 7.3B | INT4 Q8 FP16 | 320 tok/s 228 tok/s 139 tok/s | ~259 ms ~261 ms ~263 ms | |
#219 | Llama 3.2 1B | 1B | INT4 Q8 FP16 | 572 tok/s 521 tok/s 434 tok/s | ~43 ms ~43 ms ~43 ms | |
#220 | Gemma 2 2B | 2B | INT4 Q8 FP16 | 504 tok/s 430 tok/s 324 tok/s | ~80 ms ~80 ms ~81 ms | |
#221 | Phi-3-medium | 14B | INT4 Q8 FP16 | 218 tok/s 143 tok/s 81 tok/s | ~489 ms ~492 ms ~497 ms | |
#222 | OLMo 3 7B Instruct | 7B | INT4 Q8 FP16 | 260 tok/s 200 tok/s 131 tok/s | ~250 ms ~251 ms ~254 ms | |
#223 | Yi-34B | 34B | INT4 Q8 FP16 | 91 tok/s 60 tok/s 34 tok/s | ~1.2s ~1.2s ~1.2s | |
#224 | Qwen2-7B | 7B | INT4 Q8 FP16 | 326 tok/s 234 tok/s 144 tok/s | ~249 ms ~251 ms ~253 ms | |
#225 | Phi-3-small | 7B | INT4 Q8 FP16 | 326 tok/s 234 tok/s 144 tok/s | ~249 ms ~251 ms ~253 ms | |
#226 | Gemma 3 1B | 1B | INT4 Q8 FP16 | 572 tok/s 521 tok/s 434 tok/s | ~43 ms ~43 ms ~43 ms | |
#227 | Mistral-7B-Instruct-v0.2 | 7.3B | INT4 Q8 FP16 | 320 tok/s 228 tok/s 139 tok/s | ~259 ms ~261 ms ~263 ms | |
#228 | Falcon-180B | 180B | INT4 | 26 tok/s | ~6.0s | |
#229 | Gemma 1 7B | 7B | INT4 Q8 FP16 | 260 tok/s 200 tok/s 131 tok/s | ~250 ms ~251 ms ~254 ms | |
#230 | Gemma 1 2B | 2B | INT4 Q8 FP16 | 518 tok/s 439 tok/s 329 tok/s | ~80 ms ~80 ms ~81 ms | |
#231 | ChatGLM3-6B | 6B | INT4 Q8 FP16 | 280 tok/s 220 tok/s 146 tok/s | ~216 ms ~217 ms ~219 ms | |
#232 | ChatGLM2-6B | 6B | INT4 Q8 FP16 | 280 tok/s 220 tok/s 146 tok/s | ~216 ms ~217 ms ~219 ms | |
#233 | ChatGLM-6B | 6B | INT4 Q8 FP16 | 280 tok/s 220 tok/s 146 tok/s | ~216 ms ~217 ms ~219 ms | |
Unranked Models (Sorted by Speed) | ||||||
- | Qwen2-0.5B | 0.5B | INT4 Q8 FP16 | 618 tok/s 587 tok/s 528 tok/s | ~22 ms ~22 ms ~22 ms | |
- | Qwen2.5-0.5B | 0.5B | INT4 Q8 FP16 | 618 tok/s 587 tok/s 528 tok/s | ~22 ms ~22 ms ~22 ms | |
- | ERNIE-4.5-0.3B | 0.3B | INT4 Q8 FP16 | 614 tok/s 597 tok/s 559 tok/s | ~15 ms ~15 ms ~15 ms | |
- | ERNIE-4.5-0.3B-Base | 0.3B | INT4 Q8 FP16 | 614 tok/s 597 tok/s 559 tok/s | ~15 ms ~15 ms ~15 ms | |
- | Falcon-1B | 1B | INT4 Q8 FP16 | 582 tok/s 529 tok/s 440 tok/s | ~43 ms ~43 ms ~43 ms | |
- | Falcon3-1B | 1B | INT4 Q8 FP16 | 572 tok/s 521 tok/s 434 tok/s | ~43 ms ~43 ms ~43 ms | |
- | Qwen2-1.5B | 1.5B | INT4 Q8 FP16 | 536 tok/s 471 tok/s 371 tok/s | ~61 ms ~61 ms ~62 ms | |
- | Qwen2.5-1.5B | 1.5B | INT4 Q8 FP16 | 536 tok/s 471 tok/s 371 tok/s | ~61 ms ~61 ms ~62 ms | |
- | Spark X2.5 1.7B | 1.7B | INT4 Q8 FP16 | 523 tok/s 454 tok/s 350 tok/s | ~69 ms ~69 ms ~70 ms | |
- | Hy-MT2 1.8B | 1.8B | INT4 Q8 FP16 | 517 tok/s 446 tok/s 341 tok/s | ~72 ms ~72 ms ~73 ms | |
- | Falcon-3B | 3B | INT4 Q8 FP16 | 467 tok/s 376 tok/s 263 tok/s | ~113 ms ~114 ms ~115 ms | |
- | CroissantLLM Base | 1.3B | INT4 Q8 FP16 | 463 tok/s 425 tok/s 351 tok/s | ~54 ms ~54 ms ~55 ms | |
- | Phi-1 | 1.3B | INT4 Q8 FP16 | 463 tok/s 425 tok/s 351 tok/s | ~54 ms ~54 ms ~55 ms | |
- | Phi-1.5 | 1.3B | INT4 Q8 FP16 | 463 tok/s 425 tok/s 351 tok/s | ~54 ms ~54 ms ~55 ms | |
- | DeepSeek-R1 3B | 3B | INT4 Q8 FP16 | 458 tok/s 370 tok/s 260 tok/s | ~113 ms ~114 ms ~115 ms | |
- | Falcon3-3B | 3B | INT4 Q8 FP16 | 455 tok/s 368 tok/s 259 tok/s | ~113 ms ~114 ms ~115 ms | |
- | Qwen2.5-3B | 3B | INT4 Q8 FP16 | 455 tok/s 368 tok/s 259 tok/s | ~113 ms ~114 ms ~115 ms | |
- | ERNIE-4.5-21B-A3B | 21B (3B active) | INT4 Q8 FP16 | 425 tok/s 350 tok/s 250 tok/s | ~138 ms ~139 ms ~140 ms | |
- | ERNIE-4.5-21B-A3B-Base | 21B (3B active) | INT4 Q8 FP16 | 425 tok/s 350 tok/s 250 tok/s | ~138 ms ~139 ms ~140 ms | |
- | ERNIE-4.5-VL-28B-A3B | 28B (3B active) | INT4 Q8 FP16 | 417 tok/s 345 tok/s 247 tok/s | ~145 ms ~146 ms ~147 ms | |
- | ERNIE-4.5-VL-28B-A3B-Base | 28B (3B active) | INT4 Q8 FP16 | 417 tok/s 345 tok/s 247 tok/s | ~145 ms ~146 ms ~147 ms | |
- | Hy-MT2 30B A3B | 30B (3B active) | INT4 Q8 FP16 | 414 tok/s 343 tok/s 246 tok/s | ~147 ms ~148 ms ~149 ms | |
- | Spark X2.5 4B | 4B | INT4 Q8 FP16 | 414 tok/s 322 tok/s 216 tok/s | ~147 ms ~148 ms ~150 ms | |
- | Phi-2 | 2.7B | INT4 Q8 FP16 | 377 tok/s 325 tok/s 243 tok/s | ~104 ms ~104 ms ~105 ms | |
- | MaLLaM-3B | 3B | INT4 Q8 FP16 | 366 tok/s 312 tok/s 230 tok/s | ~114 ms ~114 ms ~116 ms | |
- | SmolLM3 3B | 3B | INT4 Q8 FP16 | 366 tok/s 312 tok/s 230 tok/s | ~114 ms ~114 ms ~116 ms | |
- | AliceAI Foundation 80B A3B Base | 80B (3B active) | INT4 Q8 | 363 tok/s 310 tok/s | ~201 ms ~202 ms | |
- | Falcon-7B | 7B | INT4 Q8 FP16 | 336 tok/s 238 tok/s 146 tok/s | ~249 ms ~251 ms ~253 ms | |
- | Falcon3-7B | 7B | INT4 Q8 FP16 | 326 tok/s 234 tok/s 144 tok/s | ~249 ms ~251 ms ~253 ms | |
- | Hy-MT2 7B | 7B | INT4 Q8 FP16 | 326 tok/s 234 tok/s 144 tok/s | ~249 ms ~251 ms ~253 ms | |
- | Mistral-7B-v0.1 | 7.3B | INT4 Q8 FP16 | 320 tok/s 228 tok/s 139 tok/s | ~259 ms ~261 ms ~263 ms | |
- | LensVLM 9B | 9B | INT4 Q8 FP16 | 286 tok/s 198 tok/s 118 tok/s | ~318 ms ~320 ms ~323 ms | |
- | MiMo V2.6 Distill Qwen 9B | 9B | INT4 Q8 FP16 | 286 tok/s 198 tok/s 118 tok/s | ~318 ms ~320 ms ~323 ms | |
- | ChatGLM3-6B-32K | 6B | INT4 Q8 FP16 | 280 tok/s 220 tok/s 146 tok/s | ~216 ms ~217 ms ~219 ms | |
- | Yi-6B | 6B | INT4 Q8 FP16 | 280 tok/s 220 tok/s 146 tok/s | ~216 ms ~217 ms ~219 ms | |
- | Kimi-VL-A3B-Instruct | 16B (3B active) | INT4 Q8 FP16 | 270 tok/s 244 tok/s 191 tok/s | ~134 ms ~134 ms ~135 ms | |
- | Falcon3-10B | 10B | INT4 Q8 FP16 | 269 tok/s 184 tok/s 108 tok/s | ~352 ms ~354 ms ~358 ms | |
- | Falcon2-11B | 11B | INT4 Q8 FP16 | 262 tok/s 175 tok/s 101 tok/s | ~386 ms ~388 ms ~392 ms | |
- | Hunyuan Lite | 7B | INT4 Q8 FP16 | 260 tok/s 200 tok/s 131 tok/s | ~250 ms ~251 ms ~254 ms | |
- | OLMo 3 7B Base | 7B | INT4 Q8 FP16 | 260 tok/s 200 tok/s 131 tok/s | ~250 ms ~251 ms ~254 ms | |
- | SEA-LION-7B | 7.1B | INT4 Q8 FP16 | 259 tok/s 199 tok/s 129 tok/s | ~254 ms ~255 ms ~257 ms | |
- | SEA-LION-7B-Instruct | 7.1B | INT4 Q8 FP16 | 259 tok/s 199 tok/s 129 tok/s | ~254 ms ~255 ms ~257 ms | |
- | Sahabat-AI-Llama3-8B-Instruct | 8B | INT4 Q8 FP16 | 243 tok/s 184 tok/s 118 tok/s | ~284 ms ~285 ms ~289 ms | |
- | Typhoon-2-8B | 8B | INT4 Q8 FP16 | 243 tok/s 184 tok/s 118 tok/s | ~284 ms ~285 ms ~289 ms | |
- | Laguna S 2.1 | 118B (8B active) | INT4 Q8 | 242 tok/s 174 tok/s | ~418 ms ~455 ms | |
- | GLM-4-9B | 9B | INT4 Q8 FP16 | 226 tok/s 169 tok/s 107 tok/s | ~319 ms ~320 ms ~324 ms | |
- | GLM-4-9B-Chat | 9B | INT4 Q8 FP16 | 226 tok/s 169 tok/s 107 tok/s | ~319 ms ~320 ms ~324 ms | |
- | GLM-4-9B-Chat-1M | 9B | INT4 Q8 FP16 | 226 tok/s 169 tok/s 107 tok/s | ~319 ms ~320 ms ~324 ms | |
- | Yi-9B | 9B | INT4 Q8 FP16 | 226 tok/s 169 tok/s 107 tok/s | ~319 ms ~320 ms ~324 ms | |
- | Sahabat-AI-Gemma2-9B | 9.2B | INT4 Q8 FP16 | 224 tok/s 167 tok/s 105 tok/s | ~326 ms ~327 ms ~331 ms | |
- | Sahabat-AI-Gemma2-9B-Instruct | 9.2B | INT4 Q8 FP16 | 224 tok/s 167 tok/s 105 tok/s | ~326 ms ~327 ms ~331 ms | |
- | GreenMind-14B-R1 | 14B | INT4 Q8 FP16 | 170 tok/s 122 tok/s 74 tok/s | ~490 ms ~493 ms ~498 ms | |
- | Hunyuan A13B | 80B (13B active) | INT4 Q8 | 108 tok/s 90 tok/s | ~532 ms ~534 ms | |
- | Falcon-40B | 40B | INT4 Q8 FP16 | 101 tok/s 59 tok/s 31 tok/s | ~1.4s ~1.4s ~1.4s | |
- | Typhoon-2-70B | 70B | INT4 Q8 | 49 tok/s 31 tok/s | ~2.4s ~2.4s | |
- | Kimi-Dev-72B | 72B | INT4 Q8 | 48 tok/s 30 tok/s | ~2.4s ~2.5s | |
- | GLM-130B | 130B | INT4 | 27 tok/s | ~4.4s | |
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 Instinct MI250 (128GB).
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 3276 GB/s aggregate bandwidth, batch size 1 inference operates in a memory-bandwidth bound regime. Generates approximately 728 tok/s on an 8B Q4 model and 86 tok/s on a 70B Q4 model.
Frequently Asked Questions
©2025 ApX Machine Learning