Instinct MI210 (64GB)
The Instinct MI210 (64GB) provides 64 GB of memory with 1638 GB/s bandwidth, supporting local inference via ROCm and Vulkan acceleration.
Memory
64 GB
Usable Memory Ceiling
64 GB
Bandwidth
1638 GB/s
Max Inference Class
70B+ Models (Q4)
TDP
300W
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 MI210 (64GB)
Inference throughput, memory compatibility, and generation speeds for open-weights LLMs on Instinct MI210 (64GB).
Sort by:
Context:
KV Cache:
GPUs:
| Rank | Model | Parameters | Precision | TPS (Approx.) | TTFT | Can Run |
|---|---|---|---|---|---|---|
#33 | Qwen 3.8 27B | 27B | INT4 Q8 FP16 | 75 tok/s 44 tok/s 22 tok/s | ~1.8s ~1.9s ~2.0s | |
#48 | Agnes 3.0 Flash | 33B | INT4 Q8 FP16 | 63 tok/s 37 tok/s 17 tok/s | ~2.2s ~2.3s ~2.7s | |
#61 | Sarvam-105B | 106B (10.3B active) | INT4 | 155 tok/s | ~1.0s | |
#72 | K2 Horizon MoVA 36B A4B | 36B (4B active) | INT4 Q8 | 263 tok/s 195 tok/s | ~364 ms ~365 ms | |
#75 | Ling 3.0 Flash VL | 124B (5.5B active) | INT4 | 217 tok/s | ~793 ms | |
#83 | Qwen3.5-27B | 27B | INT4 Q8 FP16 | 75 tok/s 44 tok/s 22 tok/s | ~1.8s ~1.9s ~2.0s | |
#84 | Qwen3.5-122B-A10B | 122B (10B active) | INT4 | 118 tok/s | ~1.1s | |
#92 | Sarvam-30B | 32B (2.4B active) | INT4 Q8 FP16 | 323 tok/s 257 tok/s 157 tok/s | ~254 ms ~255 ms ~302 ms | |
#94 | Qwen3.6 35B A3B | 35B (3B active) | INT4 Q8 | 297 tok/s 229 tok/s | ~297 ms ~298 ms | |
#96 | Muse Glimmer 30B | 30B | INT4 Q8 FP16 | 69 tok/s 40 tok/s 19 tok/s | ~2.0s ~2.1s ~2.4s | |
#100 | Phi-4 Reasoning Plus | 14B | INT4 Q8 FP16 | 98 tok/s 67 tok/s 39 tok/s | ~976 ms ~981 ms ~991 ms | |
#101 | MiMo V2 Flash | 15B (309B active) | INT4 Q8 FP16 | 8 tok/s 4 tok/s 2 tok/s | ~20.2s ~20.3s ~20.5s | |
#108 | Gemma 4 26B A4B | 25.2B (3.8B active) | INT4 Q8 FP16 | 278 tok/s 205 tok/s 122 tok/s | ~333 ms ~334 ms ~366 ms | |
#109 | Gemma 4 12B | 11.95B | INT4 Q8 FP16 | 110 tok/s 77 tok/s 45 tok/s | ~836 ms ~840 ms ~849 ms | |
#113 | Qwen3.5-9B | 9B | INT4 Q8 FP16 | 182 tok/s 116 tok/s 65 tok/s | ~631 ms ~635 ms ~641 ms | |
#114 | Qwen3.5-35B-A3B | 35B (3B active) | INT4 Q8 | 297 tok/s 229 tok/s | ~297 ms ~298 ms | |
#114 | Ling 3.0 Flash | 124B (5.1B active) | INT4 | 226 tok/s | ~762 ms | |
#118 | K2 Horizon 7B | 7B | INT4 Q8 FP16 | 216 tok/s 142 tok/s 81 tok/s | ~495 ms ~497 ms ~502 ms | |
#126 | Qwen3 Next 80B A3B | 80B (79B active) | INT4 | 21 tok/s | ~5.8s | |
#128 | NVIDIA Nemotron 3 Nano 30B-A3B | 3.5B (30B active) | INT4 Q8 FP16 | 66 tok/s 39 tok/s 21 tok/s | ~2.0s ~2.0s ~2.0s | |
#129 | Gemma 4 31B | 30.7B | INT4 Q8 FP16 | 67 tok/s 39 tok/s 18 tok/s | ~2.1s ~2.1s ~2.5s | |
#131 | Nemotron 3.5 Lightning | 30B (3B active) | INT4 Q8 FP16 | 135 tok/s 124 tok/s 87 tok/s | ~295 ms ~295 ms ~349 ms | |
#132 | Qwen3.5-4B | 4B | INT4 Q8 FP16 | 300 tok/s 212 tok/s 129 tok/s | ~291 ms ~292 ms ~295 ms | |
#133 | Qwen3-30B-A3B | 30B (3B active) | INT4 Q8 FP16 | 300 tok/s 231 tok/s 136 tok/s | ~291 ms ~292 ms ~345 ms | |
#136 | GPT-OSS 20B | 21B (3.6B active) | INT4 Q8 FP16 | 149 tok/s 130 tok/s 96 tok/s | ~315 ms ~316 ms ~318 ms | |
#146 | Granite 4.2 8B | 8B | INT4 Q8 FP16 | 198 tok/s 128 tok/s 72 tok/s | ~562 ms ~565 ms ~571 ms | |
#150 | Gemma 4 E4B | 8B | INT4 Q8 FP16 | 198 tok/s 128 tok/s 72 tok/s | ~562 ms ~565 ms ~571 ms | |
#151 | Nemotron 3.5 Lightning 30B A3B | 30B (3B active) | INT4 Q8 FP16 | 300 tok/s 231 tok/s 136 tok/s | ~291 ms ~292 ms ~345 ms | |
#153 | Qwen3-32B | 32B | INT4 Q8 FP16 | 65 tok/s 38 tok/s 18 tok/s | ~2.2s ~2.2s ~2.6s | |
#154 | Qwen2.5-72B | 72B | INT4 | 31 tok/s | ~4.9s | |
#155 | Mistral Large 3 | 41B (675B active) | INT4 Q8 | 3 tok/s 2 tok/s | ~44.1s ~48.2s | |
#156 | GLM-4 | 32B | INT4 Q8 | 51 tok/s 33 tok/s | ~2.2s ~2.2s | |
#157 | DeepSeek-R1 70B | 70B | INT4 | 32 tok/s | ~4.7s | |
#158 | GLM-4V | 9B | INT4 Q8 FP16 | 136 tok/s 97 tok/s 58 tok/s | ~633 ms ~636 ms ~643 ms | |
#159 | Gemma 3 27B | 27B | INT4 Q8 FP16 | 75 tok/s 44 tok/s 22 tok/s | ~1.8s ~1.9s ~2.0s | |
#160 | Phi-4 | 14B | INT4 Q8 FP16 | 130 tok/s 80 tok/s 43 tok/s | ~973 ms ~978 ms ~989 ms | |
#161 | Llama 3.3 70B | 70B | INT4 | 32 tok/s | ~4.7s | |
#162 | DeepSeek-R1 32B | 32B | INT4 Q8 FP16 | 67 tok/s 38 tok/s 18 tok/s | ~2.2s ~2.2s ~2.6s | |
#163 | Hunyuan Standard | 52B (389B active) | INT4 Q8 | 6 tok/s 3 tok/s | ~25.5s ~30.1s | |
#164 | Qwen2.5-32B | 32B | INT4 Q8 FP16 | 65 tok/s 38 tok/s 18 tok/s | ~2.2s ~2.2s ~2.6s | |
#168 | Magistral Small | 24B | INT4 Q8 FP16 | 64 tok/s 42 tok/s 23 tok/s | ~1.7s ~1.7s ~1.8s | |
#169 | OLMo 3 7B Think | 7B | INT4 Q8 FP16 | 162 tok/s 118 tok/s 72 tok/s | ~496 ms ~499 ms ~504 ms | |
#170 | OLMo 3.1 32B Instruct | 32B | INT4 Q8 | 51 tok/s 33 tok/s | ~2.2s ~2.2s | |
#171 | Llama 3.1 70B | 70B | INT4 | 32 tok/s | ~4.7s | |
#172 | DeepSeek-R1 14B | 14B | INT4 Q8 FP16 | 134 tok/s 81 tok/s 43 tok/s | ~973 ms ~978 ms ~989 ms | |
#173 | Ministral 3 14B | 14B | INT4 Q8 FP16 | 98 tok/s 67 tok/s 39 tok/s | ~976 ms ~981 ms ~991 ms | |
#174 | Mistral-Large-2407 | 123B | INT4 | 17 tok/s | ~9.8s | |
#175 | Llama 4 Scout | 109B (17B active) | INT4 | 92 tok/s | ~1.5s | |
#177 | Gemma 3 12B | 12B | INT4 Q8 FP16 | 147 tok/s 91 tok/s 50 tok/s | ~837 ms ~841 ms ~850 ms | |
#179 | Qwen2-72B | 72B | INT4 | 31 tok/s | ~4.9s | |
#180 | OLMo 3 32B Base | 32B | INT4 Q8 | 51 tok/s 33 tok/s | ~2.2s ~2.2s | |
#181 | Mistral-Small-2501 | 24B | INT4 Q8 FP16 | 84 tok/s 49 tok/s 26 tok/s | ~1.6s ~1.7s ~1.7s | |
#182 | Ministral 3 8B | 8B | INT4 Q8 FP16 | 149 tok/s 107 tok/s 65 tok/s | ~564 ms ~567 ms ~573 ms | |
#183 | Qwen3.5-2B | 2B | INT4 Q8 FP16 | 405 tok/s 317 tok/s 214 tok/s | ~156 ms ~156 ms ~158 ms | |
#184 | MiniCPM5-2B | 2B | INT4 Q8 FP16 | 405 tok/s 317 tok/s 214 tok/s | ~156 ms ~156 ms ~158 ms | |
#185 | Qwen2.5-14B | 14B | INT4 Q8 FP16 | 130 tok/s 80 tok/s 43 tok/s | ~973 ms ~978 ms ~989 ms | |
#186 | Gemma 2 27B | 27B | INT4 Q8 FP16 | 75 tok/s 44 tok/s 22 tok/s | ~1.8s ~1.9s ~2.0s | |
#187 | OLMo 3.1 32B Think | 32B | INT4 Q8 | 51 tok/s 33 tok/s | ~2.2s ~2.2s | |
#188 | Gemma 4 E2B | 5.1B | INT4 Q8 FP16 | 263 tok/s 180 tok/s 106 tok/s | ~366 ms ~367 ms ~371 ms | |
#189 | Qwen3-4B | 4B | INT4 Q8 FP16 | 300 tok/s 212 tok/s 129 tok/s | ~291 ms ~292 ms ~295 ms | |
#190 | DeepSeek-R1 8B | 8B | INT4 Q8 FP16 | 202 tok/s 129 tok/s 72 tok/s | ~562 ms ~565 ms ~571 ms | |
#191 | Llama 3 70B | 70B | INT4 | 32 tok/s | ~4.7s | |
#193 | Ministral 3 3B | 3B | INT4 Q8 FP16 | 252 tok/s 203 tok/s 139 tok/s | ~224 ms ~225 ms ~227 ms | |
#195 | Qwen3-14B | 14B | INT4 Q8 FP16 | 130 tok/s 80 tok/s 43 tok/s | ~973 ms ~978 ms ~989 ms | |
#196 | Kimi Linear 48B A3B Instruct | 48B (3B active) | INT4 Q8 | 116 tok/s 104 tok/s | ~319 ms ~347 ms | |
#197 | Gemma 2 9B | 9B | INT4 Q8 FP16 | 182 tok/s 116 tok/s 65 tok/s | ~631 ms ~635 ms ~641 ms | |
#198 | Qwen2.5-7B | 7B | INT4 Q8 FP16 | 216 tok/s 142 tok/s 81 tok/s | ~495 ms ~497 ms ~502 ms | |
#199 | Qwen3-8B | 8B | INT4 Q8 FP16 | 198 tok/s 128 tok/s 72 tok/s | ~562 ms ~565 ms ~571 ms | |
#200 | Gemma 3 4B | 4B | INT4 Q8 FP16 | 300 tok/s 212 tok/s 129 tok/s | ~291 ms ~292 ms ~295 ms | |
#201 | DeepSeek-R1 7B | 7B | INT4 Q8 FP16 | 221 tok/s 144 tok/s 81 tok/s | ~495 ms ~497 ms ~502 ms | |
#202 | Llama 3.1 8B | 8B | INT4 Q8 FP16 | 198 tok/s 128 tok/s 72 tok/s | ~562 ms ~565 ms ~571 ms | |
#204 | Qwen3-1.7B | 1.7B | INT4 Q8 FP16 | 430 tok/s 344 tok/s 238 tok/s | ~134 ms ~134 ms ~135 ms | |
#205 | Command R | 35B | INT4 Q8 | 48 tok/s 31 tok/s | ~2.4s ~2.4s | |
#206 | Gemma 3 270M | 0.27B | INT4 Q8 FP16 | 580 tok/s 552 tok/s 496 tok/s | ~23 ms ~23 ms ~24 ms | |
#208 | Llama 3 8B | 8B | INT4 Q8 FP16 | 198 tok/s 128 tok/s 72 tok/s | ~562 ms ~565 ms ~571 ms | |
#209 | Qwen3-0.6B | 0.6B | INT4 Q8 FP16 | 559 tok/s 502 tok/s 408 tok/s | ~49 ms ~49 ms ~49 ms | |
#210 | Phi-4-Mini | 3.8B | INT4 Q8 FP16 | 308 tok/s 220 tok/s 134 tok/s | ~277 ms ~279 ms ~282 ms | |
#211 | Mixtral-8x7B-v0.1 | 46.7B (7B active) | INT4 Q8 | 195 tok/s 133 tok/s | ~574 ms ~576 ms | |
#212 | Ministral-8B-2410 | 8B | INT4 Q8 FP16 | 198 tok/s 128 tok/s 72 tok/s | ~562 ms ~565 ms ~571 ms | |
#213 | DeepSeek-R1 1.5B | 1.5B | INT4 Q8 FP16 | 448 tok/s 364 tok/s 257 tok/s | ~118 ms ~119 ms ~120 ms | |
#214 | Llama 3.2 3B | 3B | INT4 Q8 FP16 | 345 tok/s 254 tok/s 161 tok/s | ~223 ms ~224 ms ~226 ms | |
#215 | Qwen3.5-0.8B | 0.8B | INT4 Q8 FP16 | 529 tok/s 462 tok/s 360 tok/s | ~65 ms ~65 ms ~66 ms | |
#216 | Phi-3-mini | 3.8B | INT4 Q8 FP16 | 308 tok/s 220 tok/s 134 tok/s | ~277 ms ~279 ms ~282 ms | |
#217 | Gemma 3n E2B IT | 6B (2B active) | INT4 Q8 FP16 | 252 tok/s 220 tok/s 165 tok/s | ~168 ms ~168 ms ~170 ms | |
#218 | Mistral-7B-Instruct-v0.1 | 7.3B | INT4 Q8 FP16 | 210 tok/s 137 tok/s 78 tok/s | ~515 ms ~517 ms ~523 ms | |
#219 | Llama 3.2 1B | 1B | INT4 Q8 FP16 | 500 tok/s 427 tok/s 322 tok/s | ~82 ms ~82 ms ~83 ms | |
#220 | Gemma 2 2B | 2B | INT4 Q8 FP16 | 405 tok/s 317 tok/s 214 tok/s | ~156 ms ~156 ms ~158 ms | |
#221 | Phi-3-medium | 14B | INT4 Q8 FP16 | 130 tok/s 80 tok/s 43 tok/s | ~973 ms ~978 ms ~989 ms | |
#222 | OLMo 3 7B Instruct | 7B | INT4 Q8 FP16 | 162 tok/s 118 tok/s 72 tok/s | ~496 ms ~499 ms ~504 ms | |
#223 | Yi-34B | 34B | INT4 Q8 | 49 tok/s 31 tok/s | ~2.3s ~2.3s | |
#224 | Qwen2-7B | 7B | INT4 Q8 FP16 | 216 tok/s 142 tok/s 81 tok/s | ~495 ms ~497 ms ~502 ms | |
#225 | Phi-3-small | 7B | INT4 Q8 FP16 | 216 tok/s 142 tok/s 81 tok/s | ~495 ms ~497 ms ~502 ms | |
#226 | Gemma 3 1B | 1B | INT4 Q8 FP16 | 500 tok/s 427 tok/s 322 tok/s | ~82 ms ~82 ms ~83 ms | |
#227 | Mistral-7B-Instruct-v0.2 | 7.3B | INT4 Q8 FP16 | 210 tok/s 137 tok/s 78 tok/s | ~515 ms ~517 ms ~523 ms | |
#229 | Gemma 1 7B | 7B | INT4 Q8 FP16 | 162 tok/s 118 tok/s 72 tok/s | ~496 ms ~499 ms ~504 ms | |
#230 | Gemma 1 2B | 2B | INT4 Q8 FP16 | 423 tok/s 327 tok/s 218 tok/s | ~156 ms ~156 ms ~158 ms | |
#231 | ChatGLM3-6B | 6B | INT4 Q8 FP16 | 178 tok/s 132 tok/s 82 tok/s | ~428 ms ~430 ms ~435 ms | |
#232 | ChatGLM2-6B | 6B | INT4 Q8 FP16 | 178 tok/s 132 tok/s 82 tok/s | ~428 ms ~430 ms ~435 ms | |
#233 | ChatGLM-6B | 6B | INT4 Q8 FP16 | 178 tok/s 132 tok/s 82 tok/s | ~428 ms ~430 ms ~435 ms | |
Unranked Models (Sorted by Speed) | ||||||
- | Qwen2-0.5B | 0.5B | INT4 Q8 FP16 | 575 tok/s 524 tok/s 436 tok/s | ~41 ms ~41 ms ~41 ms | |
- | Qwen2.5-0.5B | 0.5B | INT4 Q8 FP16 | 575 tok/s 524 tok/s 436 tok/s | ~41 ms ~41 ms ~41 ms | |
- | ERNIE-4.5-0.3B | 0.3B | INT4 Q8 FP16 | 569 tok/s 540 tok/s 481 tok/s | ~26 ms ~26 ms ~26 ms | |
- | ERNIE-4.5-0.3B-Base | 0.3B | INT4 Q8 FP16 | 569 tok/s 540 tok/s 481 tok/s | ~26 ms ~26 ms ~26 ms | |
- | Falcon-1B | 1B | INT4 Q8 FP16 | 517 tok/s 438 tok/s 328 tok/s | ~82 ms ~82 ms ~83 ms | |
- | Falcon3-1B | 1B | INT4 Q8 FP16 | 500 tok/s 427 tok/s 322 tok/s | ~82 ms ~82 ms ~83 ms | |
- | Qwen2-1.5B | 1.5B | INT4 Q8 FP16 | 448 tok/s 364 tok/s 257 tok/s | ~118 ms ~119 ms ~120 ms | |
- | Qwen2.5-1.5B | 1.5B | INT4 Q8 FP16 | 448 tok/s 364 tok/s 257 tok/s | ~118 ms ~119 ms ~120 ms | |
- | Spark X2.5 1.7B | 1.7B | INT4 Q8 FP16 | 430 tok/s 344 tok/s 238 tok/s | ~134 ms ~134 ms ~135 ms | |
- | Hy-MT2 1.8B | 1.8B | INT4 Q8 FP16 | 422 tok/s 335 tok/s 229 tok/s | ~140 ms ~141 ms ~142 ms | |
- | Falcon-3B | 3B | INT4 Q8 FP16 | 360 tok/s 262 tok/s 164 tok/s | ~223 ms ~224 ms ~226 ms | |
- | CroissantLLM Base | 1.3B | INT4 Q8 FP16 | 355 tok/s 312 tok/s 238 tok/s | ~104 ms ~105 ms ~106 ms | |
- | Phi-1 | 1.3B | INT4 Q8 FP16 | 355 tok/s 312 tok/s 238 tok/s | ~104 ms ~105 ms ~106 ms | |
- | Phi-1.5 | 1.3B | INT4 Q8 FP16 | 355 tok/s 312 tok/s 238 tok/s | ~104 ms ~105 ms ~106 ms | |
- | DeepSeek-R1 3B | 3B | INT4 Q8 FP16 | 348 tok/s 256 tok/s 161 tok/s | ~223 ms ~224 ms ~226 ms | |
- | Falcon3-3B | 3B | INT4 Q8 FP16 | 345 tok/s 254 tok/s 161 tok/s | ~223 ms ~224 ms ~226 ms | |
- | Qwen2.5-3B | 3B | INT4 Q8 FP16 | 345 tok/s 254 tok/s 161 tok/s | ~223 ms ~224 ms ~226 ms | |
- | ERNIE-4.5-21B-A3B | 21B (3B active) | INT4 Q8 FP16 | 311 tok/s 237 tok/s 154 tok/s | ~272 ms ~273 ms ~276 ms | |
- | ERNIE-4.5-21B-A3B-Base | 21B (3B active) | INT4 Q8 FP16 | 311 tok/s 237 tok/s 154 tok/s | ~272 ms ~273 ms ~276 ms | |
- | ERNIE-4.5-VL-28B-A3B | 28B (3B active) | INT4 Q8 FP16 | 303 tok/s 233 tok/s 144 tok/s | ~286 ms ~287 ms ~314 ms | |
- | ERNIE-4.5-VL-28B-A3B-Base | 28B (3B active) | INT4 Q8 FP16 | 303 tok/s 233 tok/s 144 tok/s | ~286 ms ~287 ms ~314 ms | |
- | Hy-MT2 30B A3B | 30B (3B active) | INT4 Q8 FP16 | 300 tok/s 231 tok/s 136 tok/s | ~291 ms ~292 ms ~345 ms | |
- | Spark X2.5 4B | 4B | INT4 Q8 FP16 | 300 tok/s 212 tok/s 129 tok/s | ~291 ms ~292 ms ~295 ms | |
- | Phi-2 | 2.7B | INT4 Q8 FP16 | 262 tok/s 215 tok/s 149 tok/s | ~204 ms ~205 ms ~207 ms | |
- | MaLLaM-3B | 3B | INT4 Q8 FP16 | 252 tok/s 203 tok/s 139 tok/s | ~224 ms ~225 ms ~227 ms | |
- | SmolLM3 3B | 3B | INT4 Q8 FP16 | 252 tok/s 203 tok/s 139 tok/s | ~224 ms ~225 ms ~227 ms | |
- | AliceAI Foundation 80B A3B Base | 80B (3B active) | INT4 | 249 tok/s | ~398 ms | |
- | Falcon-7B | 7B | INT4 Q8 FP16 | 224 tok/s 145 tok/s 82 tok/s | ~495 ms ~497 ms ~502 ms | |
- | Falcon3-7B | 7B | INT4 Q8 FP16 | 216 tok/s 142 tok/s 81 tok/s | ~495 ms ~497 ms ~502 ms | |
- | Hy-MT2 7B | 7B | INT4 Q8 FP16 | 216 tok/s 142 tok/s 81 tok/s | ~495 ms ~497 ms ~502 ms | |
- | Mistral-7B-v0.1 | 7.3B | INT4 Q8 FP16 | 210 tok/s 137 tok/s 78 tok/s | ~515 ms ~517 ms ~523 ms | |
- | LensVLM 9B | 9B | INT4 Q8 FP16 | 182 tok/s 116 tok/s 65 tok/s | ~631 ms ~635 ms ~641 ms | |
- | MiMo V2.6 Distill Qwen 9B | 9B | INT4 Q8 FP16 | 182 tok/s 116 tok/s 65 tok/s | ~631 ms ~635 ms ~641 ms | |
- | ChatGLM3-6B-32K | 6B | INT4 Q8 FP16 | 178 tok/s 132 tok/s 82 tok/s | ~428 ms ~430 ms ~435 ms | |
- | Yi-6B | 6B | INT4 Q8 FP16 | 178 tok/s 132 tok/s 82 tok/s | ~428 ms ~430 ms ~435 ms | |
- | Falcon3-10B | 10B | INT4 Q8 FP16 | 169 tok/s 106 tok/s 59 tok/s | ~699 ms ~703 ms ~710 ms | |
- | Kimi-VL-A3B-Instruct | 16B (3B active) | INT4 Q8 FP16 | 169 tok/s 149 tok/s 111 tok/s | ~264 ms ~265 ms ~267 ms | |
- | Falcon2-11B | 11B | INT4 Q8 FP16 | 163 tok/s 100 tok/s 54 tok/s | ~767 ms ~771 ms ~780 ms | |
- | Hunyuan Lite | 7B | INT4 Q8 FP16 | 162 tok/s 118 tok/s 72 tok/s | ~496 ms ~499 ms ~504 ms | |
- | OLMo 3 7B Base | 7B | INT4 Q8 FP16 | 162 tok/s 118 tok/s 72 tok/s | ~496 ms ~499 ms ~504 ms | |
- | SEA-LION-7B | 7.1B | INT4 Q8 FP16 | 160 tok/s 117 tok/s 72 tok/s | ~503 ms ~505 ms ~511 ms | |
- | SEA-LION-7B-Instruct | 7.1B | INT4 Q8 FP16 | 160 tok/s 117 tok/s 72 tok/s | ~503 ms ~505 ms ~511 ms | |
- | Sahabat-AI-Llama3-8B-Instruct | 8B | INT4 Q8 FP16 | 149 tok/s 107 tok/s 65 tok/s | ~564 ms ~567 ms ~573 ms | |
- | Typhoon-2-8B | 8B | INT4 Q8 FP16 | 149 tok/s 107 tok/s 65 tok/s | ~564 ms ~567 ms ~573 ms | |
- | GLM-4-9B | 9B | INT4 Q8 FP16 | 136 tok/s 97 tok/s 58 tok/s | ~633 ms ~636 ms ~643 ms | |
- | GLM-4-9B-Chat | 9B | INT4 Q8 FP16 | 136 tok/s 97 tok/s 58 tok/s | ~633 ms ~636 ms ~643 ms | |
- | GLM-4-9B-Chat-1M | 9B | INT4 Q8 FP16 | 136 tok/s 97 tok/s 58 tok/s | ~633 ms ~636 ms ~643 ms | |
- | Yi-9B | 9B | INT4 Q8 FP16 | 136 tok/s 97 tok/s 58 tok/s | ~633 ms ~636 ms ~643 ms | |
- | Sahabat-AI-Gemma2-9B | 9.2B | INT4 Q8 FP16 | 134 tok/s 95 tok/s 57 tok/s | ~647 ms ~650 ms ~657 ms | |
- | Sahabat-AI-Gemma2-9B-Instruct | 9.2B | INT4 Q8 FP16 | 134 tok/s 95 tok/s 57 tok/s | ~647 ms ~650 ms ~657 ms | |
- | Laguna S 2.1 | 118B (8B active) | INT4 | 133 tok/s | ~978 ms | |
- | GreenMind-14B-R1 | 14B | INT4 Q8 FP16 | 98 tok/s 67 tok/s 39 tok/s | ~976 ms ~981 ms ~991 ms | |
- | Hunyuan A13B | 80B (13B active) | INT4 | 56 tok/s | ~1.1s | |
- | Falcon-40B | 40B | INT4 Q8 | 55 tok/s 31 tok/s | ~2.7s ~2.7s | |
- | Kimi-Dev-72B | 72B | INT4 | 25 tok/s | ~4.9s | |
- | Typhoon-2-70B | 70B | INT4 | 25 tok/s | ~4.7s | |
Inference Capacity
Maximum model parameter class runnable by weight precision.
Context:
4-bit
70B+ Models
8-bit
32B Models
16-bit
14B 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 MI210 (64GB).
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 1638 GB/s aggregate bandwidth, batch size 1 inference operates in a memory-bandwidth bound regime. Generates approximately 364 tok/s on an 8B Q4 model and 43 tok/s on a 70B Q4 model.
Frequently Asked Questions
©2025 ApX Machine Learning