NVIDIA RTX 5090 (32GB)
The NVIDIA RTX 5090 (32GB) features 32 GB of dedicated VRAM and 1792 GB/s memory bandwidth, supporting CUDA and TensorRT-LLM runtimes for full-precision and quantized inference.
Memory
32 GB
Usable Memory Ceiling
32 GB
Bandwidth
1792 GB/s
Max Inference Class
32B Models (Q4)
TDP
575W
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 RTX 5090 (32GB)
Inference throughput, memory compatibility, and generation speeds for open-weights LLMs on NVIDIA RTX 5090 (32GB).
Sort by:
Context:
KV Cache:
GPUs:
| Rank | Model | Parameters | Precision | TPS (Approx.) | TTFT | Can Run |
|---|---|---|---|---|---|---|
#36 | Qwen 3.8 27B | 27B | INT4 Q8 | 78 tok/s 46 tok/s | ~723 ms ~792 ms | |
#80 | K2 Horizon MoVA 36B A4B | 36B (4B active) | INT4 | 214 tok/s | ~147 ms | |
#86 | Qwen3.5-27B | 27B | INT4 Q8 | 78 tok/s 46 tok/s | ~723 ms ~792 ms | |
#93 | Sarvam-30B | 32B (2.4B active) | INT4 | 245 tok/s | ~104 ms | |
#99 | Muse Glimmer 30B | 30B | INT4 Q8 | 72 tok/s 39 tok/s | ~801 ms ~948 ms | |
#102 | Phi-4 Reasoning Plus | 14B | INT4 Q8 FP16 | 98 tok/s 70 tok/s 38 tok/s | ~386 ms ~390 ms ~466 ms | |
#103 | Qwen3.6 35B A3B | 35B (3B active) | INT4 | 232 tok/s | ~121 ms | |
#105 | MiMo V2 Flash | 15B (309B active) | INT4 Q8 | 9 tok/s 5 tok/s | ~7.8s ~7.9s | |
#110 | Gemma 4 26B A4B | 25.2B (3.8B active) | INT4 Q8 | 222 tok/s 169 tok/s | ~135 ms ~147 ms | |
#120 | Qwen3.5-35B-A3B | 35B (3B active) | INT4 | 232 tok/s | ~121 ms | |
#121 | Qwen3.5-9B | 9B | INT4 Q8 FP16 | 163 tok/s 113 tok/s 68 tok/s | ~251 ms ~254 ms ~260 ms | |
#127 | K2 Horizon 7B | 7B | INT4 Q8 FP16 | 186 tok/s 134 tok/s 83 tok/s | ~198 ms ~200 ms ~205 ms | |
#130 | Gemma 4 12B | 11.95B | INT4 Q8 FP16 | 108 tok/s 79 tok/s 46 tok/s | ~331 ms ~335 ms ~371 ms | |
#136 | Qwen3.5-4B | 4B | INT4 Q8 FP16 | 234 tok/s 183 tok/s 124 tok/s | ~119 ms ~120 ms ~123 ms | |
#139 | GPT-OSS 20B | 21B (3.6B active) | INT4 Q8 | 139 tok/s 118 tok/s | ~129 ms ~141 ms | |
#140 | GLM-4V | 9B | INT4 Q8 FP16 | 129 tok/s 97 tok/s 62 tok/s | ~253 ms ~255 ms ~261 ms | |
#145 | NVIDIA Nemotron 3 Nano 30B-A3B | 3.5B (30B active) | INT4 Q8 FP16 | 69 tok/s 43 tok/s 23 tok/s | ~777 ms ~786 ms ~805 ms | |
#150 | Qwen3-30B-A3B | 30B (3B active) | INT4 Q8 | 234 tok/s 175 tok/s | ~119 ms ~139 ms | |
#152 | Granite 4.2 8B | 8B | INT4 Q8 FP16 | 174 tok/s 123 tok/s 75 tok/s | ~224 ms ~227 ms ~232 ms | |
#157 | Gemma 4 E4B | 8B | INT4 Q8 FP16 | 174 tok/s 123 tok/s 75 tok/s | ~224 ms ~227 ms ~232 ms | |
#158 | Mistral Large 3 | 41B (675B active) | INT4 | 4 tok/s | ~20.1s | |
#161 | GLM-4 | 32B | INT4 | 55 tok/s | ~856 ms | |
#163 | Gemma 3 27B | 27B | INT4 Q8 | 78 tok/s 46 tok/s | ~723 ms ~792 ms | |
#164 | Phi-4 | 14B | INT4 Q8 FP16 | 125 tok/s 82 tok/s 44 tok/s | ~384 ms ~388 ms ~430 ms | |
#166 | Qwen3-32B | 32B | INT4 | 68 tok/s | ~853 ms | |
#168 | DeepSeek-R1 32B | 32B | INT4 Q8 | 70 tok/s 38 tok/s | ~852 ms ~1.0s | |
#169 | Gemma 4 31B | 30.7B | INT4 Q8 | 71 tok/s 39 tok/s | ~819 ms ~969 ms | |
#171 | Magistral Small | 24B | INT4 Q8 | 68 tok/s 44 tok/s | ~648 ms ~710 ms | |
#172 | Qwen2.5-32B | 32B | INT4 | 68 tok/s | ~853 ms | |
#174 | OLMo 3.1 32B Instruct | 32B | INT4 | 55 tok/s | ~856 ms | |
#176 | DeepSeek-R1 14B | 14B | INT4 Q8 FP16 | 128 tok/s 83 tok/s 45 tok/s | ~383 ms ~388 ms ~430 ms | |
#177 | Ministral 3 14B | 14B | INT4 Q8 FP16 | 98 tok/s 70 tok/s 38 tok/s | ~386 ms ~390 ms ~466 ms | |
#180 | Gemma 3 12B | 12B | INT4 Q8 FP16 | 138 tok/s 92 tok/s 53 tok/s | ~331 ms ~334 ms ~342 ms | |
#184 | Mistral-Small-2501 | 24B | INT4 Q8 | 85 tok/s 50 tok/s | ~645 ms ~707 ms | |
#185 | Ministral 3 8B | 8B | INT4 Q8 FP16 | 139 tok/s 106 tok/s 68 tok/s | ~226 ms ~228 ms ~233 ms | |
#186 | Gemma 2 27B | 27B | INT4 Q8 | 78 tok/s 46 tok/s | ~723 ms ~792 ms | |
#187 | OLMo 3 7B Think | 7B | INT4 Q8 FP16 | 149 tok/s 115 tok/s 75 tok/s | ~199 ms ~201 ms ~206 ms | |
#188 | Qwen3.5-2B | 2B | INT4 Q8 FP16 | 283 tok/s 243 tok/s 184 tok/s | ~67 ms ~67 ms ~68 ms | |
#189 | MiniCPM5-2B | 2B | INT4 Q8 FP16 | 283 tok/s 243 tok/s 184 tok/s | ~67 ms ~67 ms ~68 ms | |
#190 | Qwen2.5-14B | 14B | INT4 Q8 FP16 | 125 tok/s 82 tok/s 44 tok/s | ~384 ms ~388 ms ~430 ms | |
#191 | OLMo 3.1 32B Think | 32B | INT4 | 55 tok/s | ~856 ms | |
#193 | Gemma 4 E2B | 5.1B | INT4 Q8 FP16 | 214 tok/s 161 tok/s 105 tok/s | ~148 ms ~149 ms ~153 ms | |
#195 | Ministral 3 3B | 3B | INT4 Q8 FP16 | 207 tok/s 177 tok/s 131 tok/s | ~94 ms ~94 ms ~96 ms | |
#196 | DeepSeek-R1 8B | 8B | INT4 Q8 FP16 | 177 tok/s 124 tok/s 75 tok/s | ~224 ms ~227 ms ~232 ms | |
#198 | Gemma 2 9B | 9B | INT4 Q8 FP16 | 163 tok/s 113 tok/s 68 tok/s | ~251 ms ~254 ms ~260 ms | |
#199 | Nemotron 3.5 Lightning 30B A3B | 30B (3B active) | INT4 Q8 | 234 tok/s 175 tok/s | ~119 ms ~139 ms | |
#200 | Kimi Linear 48B A3B Instruct | 48B (3B active) | INT4 | 102 tok/s | ~153 ms | |
#201 | Qwen3-4B | 4B | INT4 Q8 FP16 | 234 tok/s 183 tok/s 124 tok/s | ~119 ms ~120 ms ~123 ms | |
#202 | Qwen3-14B | 14B | INT4 Q8 FP16 | 125 tok/s 82 tok/s 44 tok/s | ~384 ms ~388 ms ~430 ms | |
#203 | Gemma 3 4B | 4B | INT4 Q8 FP16 | 234 tok/s 183 tok/s 124 tok/s | ~119 ms ~120 ms ~123 ms | |
#204 | DeepSeek-R1 7B | 7B | INT4 Q8 FP16 | 188 tok/s 135 tok/s 83 tok/s | ~198 ms ~200 ms ~205 ms | |
#205 | OLMo 3 32B Base | 32B | INT4 | 55 tok/s | ~856 ms | |
#206 | Qwen3.5-0.8B | 0.8B | INT4 Q8 FP16 | 329 tok/s 306 tok/s 263 tok/s | ~31 ms ~32 ms ~32 ms | |
#207 | Llama 3.1 8B | 8B | INT4 Q8 FP16 | 174 tok/s 123 tok/s 75 tok/s | ~224 ms ~227 ms ~232 ms | |
#207 | Qwen3-8B | 8B | INT4 Q8 FP16 | 174 tok/s 123 tok/s 75 tok/s | ~224 ms ~227 ms ~232 ms | |
#210 | Nemotron 3.5 Lightning | 30B (3B active) | INT4 | 128 tok/s | ~122 ms | |
#211 | Command R | 35B | INT4 | 51 tok/s | ~934 ms | |
#212 | Llama 3 8B | 8B | INT4 Q8 FP16 | 174 tok/s 123 tok/s 75 tok/s | ~224 ms ~227 ms ~232 ms | |
#213 | Qwen3-1.7B | 1.7B | INT4 Q8 FP16 | 293 tok/s 256 tok/s 199 tok/s | ~58 ms ~58 ms ~60 ms | |
#214 | Gemma 3 270M | 0.27B | INT4 Q8 FP16 | 346 tok/s 337 tok/s 318 tok/s | ~15 ms ~16 ms ~16 ms | |
#216 | Qwen3-0.6B | 0.6B | INT4 Q8 FP16 | 339 tok/s 320 tok/s 284 tok/s | ~25 ms ~25 ms ~26 ms | |
#217 | Phi-4-Mini | 3.8B | INT4 Q8 FP16 | 238 tok/s 188 tok/s 128 tok/s | ~114 ms ~115 ms ~117 ms | |
#218 | Qwen2.5-7B | 7B | INT4 Q8 FP16 | 186 tok/s 134 tok/s 83 tok/s | ~198 ms ~200 ms ~205 ms | |
#219 | Mixtral-8x7B-v0.1 | 46.7B (7B active) | INT4 | 163 tok/s | ~247 ms | |
#220 | Ministral-8B-2410 | 8B | INT4 Q8 FP16 | 174 tok/s 123 tok/s 75 tok/s | ~224 ms ~227 ms ~232 ms | |
#221 | DeepSeek-R1 1.5B | 1.5B | INT4 Q8 FP16 | 300 tok/s 265 tok/s 210 tok/s | ~52 ms ~53 ms ~54 ms | |
#222 | Llama 3.2 3B | 3B | INT4 Q8 FP16 | 256 tok/s 209 tok/s 148 tok/s | ~93 ms ~94 ms ~96 ms | |
#223 | Phi-3-mini | 3.8B | INT4 Q8 FP16 | 238 tok/s 188 tok/s 128 tok/s | ~114 ms ~115 ms ~117 ms | |
#224 | Gemma 3n E2B IT | 6B (2B active) | INT4 Q8 FP16 | 207 tok/s 188 tok/s 151 tok/s | ~72 ms ~72 ms ~74 ms | |
#225 | Gemma 2 2B | 2B | INT4 Q8 FP16 | 283 tok/s 243 tok/s 184 tok/s | ~67 ms ~67 ms ~68 ms | |
#226 | Mistral-7B-Instruct-v0.1 | 7.3B | INT4 Q8 FP16 | 182 tok/s 130 tok/s 80 tok/s | ~206 ms ~208 ms ~213 ms | |
#227 | Llama 3.2 1B | 1B | INT4 Q8 FP16 | 320 tok/s 292 tok/s 245 tok/s | ~38 ms ~38 ms ~39 ms | |
#228 | Phi-3-medium | 14B | INT4 Q8 FP16 | 125 tok/s 82 tok/s 44 tok/s | ~384 ms ~388 ms ~430 ms | |
#229 | Yi-34B | 34B | INT4 | 52 tok/s | ~908 ms | |
#230 | OLMo 3 7B Instruct | 7B | INT4 Q8 FP16 | 149 tok/s 115 tok/s 75 tok/s | ~199 ms ~201 ms ~206 ms | |
#231 | Qwen2-7B | 7B | INT4 Q8 FP16 | 186 tok/s 134 tok/s 83 tok/s | ~198 ms ~200 ms ~205 ms | |
#232 | Phi-3-small | 7B | INT4 Q8 FP16 | 186 tok/s 134 tok/s 83 tok/s | ~198 ms ~200 ms ~205 ms | |
#233 | Mistral-7B-Instruct-v0.2 | 7.3B | INT4 Q8 FP16 | 182 tok/s 130 tok/s 80 tok/s | ~206 ms ~208 ms ~213 ms | |
#234 | Gemma 3 1B | 1B | INT4 Q8 FP16 | 320 tok/s 292 tok/s 245 tok/s | ~38 ms ~38 ms ~39 ms | |
#236 | Gemma 1 7B | 7B | INT4 Q8 FP16 | 149 tok/s 115 tok/s 75 tok/s | ~199 ms ~201 ms ~206 ms | |
#237 | Gemma 1 2B | 2B | INT4 Q8 FP16 | 291 tok/s 248 tok/s 187 tok/s | ~66 ms ~67 ms ~68 ms | |
#238 | ChatGLM3-6B | 6B | INT4 Q8 FP16 | 160 tok/s 126 tok/s 84 tok/s | ~173 ms ~175 ms ~179 ms | |
#239 | ChatGLM2-6B | 6B | INT4 Q8 FP16 | 160 tok/s 126 tok/s 84 tok/s | ~173 ms ~175 ms ~179 ms | |
#240 | ChatGLM-6B | 6B | INT4 Q8 FP16 | 160 tok/s 126 tok/s 84 tok/s | ~173 ms ~175 ms ~179 ms | |
Unranked Models (Sorted by Speed) | ||||||
- | DINOv3 ViT-L/16 | 0.3B | INT4 Q8 FP16 | 354 tok/s 344 tok/s 322 tok/s | ~16 ms ~16 ms ~17 ms | |
- | Qwen2-0.5B | 0.5B | INT4 Q8 FP16 | 344 tok/s 328 tok/s 296 tok/s | ~22 ms ~22 ms ~23 ms | |
- | Qwen2.5-0.5B | 0.5B | INT4 Q8 FP16 | 344 tok/s 328 tok/s 296 tok/s | ~22 ms ~22 ms ~23 ms | |
- | ERNIE-4.5-0.3B | 0.3B | INT4 Q8 FP16 | 342 tok/s 333 tok/s 313 tok/s | ~16 ms ~16 ms ~17 ms | |
- | ERNIE-4.5-0.3B-Base | 0.3B | INT4 Q8 FP16 | 342 tok/s 333 tok/s 313 tok/s | ~16 ms ~16 ms ~17 ms | |
- | Liquid D1-Omni 600M | 0.6B | INT4 Q8 FP16 | 339 tok/s 320 tok/s 284 tok/s | ~25 ms ~25 ms ~26 ms | |
- | Falcon-1B | 1B | INT4 Q8 FP16 | 325 tok/s 297 tok/s 248 tok/s | ~38 ms ~38 ms ~39 ms | |
- | Falcon3-1B | 1B | INT4 Q8 FP16 | 320 tok/s 292 tok/s 245 tok/s | ~38 ms ~38 ms ~39 ms | |
- | MiniCPM-V 4.7 1B | 1B | INT4 Q8 FP16 | 320 tok/s 292 tok/s 245 tok/s | ~38 ms ~38 ms ~39 ms | |
- | Qwen2-1.5B | 1.5B | INT4 Q8 FP16 | 300 tok/s 265 tok/s 210 tok/s | ~52 ms ~53 ms ~54 ms | |
- | Qwen2.5-1.5B | 1.5B | INT4 Q8 FP16 | 300 tok/s 265 tok/s 210 tok/s | ~52 ms ~53 ms ~54 ms | |
- | Spark X2.5 1.7B | 1.7B | INT4 Q8 FP16 | 293 tok/s 256 tok/s 199 tok/s | ~58 ms ~58 ms ~60 ms | |
- | Hy-MT2 1.8B | 1.8B | INT4 Q8 FP16 | 290 tok/s 251 tok/s 194 tok/s | ~61 ms ~61 ms ~62 ms | |
- | Falcon-3B | 3B | INT4 Q8 FP16 | 263 tok/s 213 tok/s 150 tok/s | ~93 ms ~93 ms ~95 ms | |
- | CroissantLLM Base | 1.3B | INT4 Q8 FP16 | 261 tok/s 240 tok/s 199 tok/s | ~47 ms ~47 ms ~48 ms | |
- | Phi-1 | 1.3B | INT4 Q8 FP16 | 261 tok/s 240 tok/s 199 tok/s | ~47 ms ~47 ms ~48 ms | |
- | Phi-1.5 | 1.3B | INT4 Q8 FP16 | 261 tok/s 240 tok/s 199 tok/s | ~47 ms ~47 ms ~48 ms | |
- | DeepSeek-R1 3B | 3B | INT4 Q8 FP16 | 258 tok/s 210 tok/s 148 tok/s | ~93 ms ~94 ms ~95 ms | |
- | Mellum 2.1 12B A2.5B Thinking | 12B (2.5B active) | INT4 Q8 FP16 | 258 tok/s 217 tok/s 160 tok/s | ~91 ms ~91 ms ~93 ms | |
- | Falcon3-3B | 3B | INT4 Q8 FP16 | 256 tok/s 209 tok/s 148 tok/s | ~93 ms ~94 ms ~96 ms | |
- | LFM2-VL 3B | 3B | INT4 Q8 FP16 | 256 tok/s 209 tok/s 148 tok/s | ~93 ms ~94 ms ~96 ms | |
- | Liquid D1 3B | 3B | INT4 Q8 FP16 | 256 tok/s 209 tok/s 148 tok/s | ~93 ms ~94 ms ~96 ms | |
- | Liquid D1-VL 3B | 3B | INT4 Q8 FP16 | 256 tok/s 209 tok/s 148 tok/s | ~93 ms ~94 ms ~96 ms | |
- | Liquid LFM2-VL 3B D1 | 3B | INT4 Q8 FP16 | 256 tok/s 209 tok/s 148 tok/s | ~93 ms ~94 ms ~96 ms | |
- | Qwen2.5-3B | 3B | INT4 Q8 FP16 | 256 tok/s 209 tok/s 148 tok/s | ~93 ms ~94 ms ~96 ms | |
- | LFM2.5 24B A2B | 24B (2.5B active) | INT4 Q8 | 248 tok/s 200 tok/s | ~101 ms ~110 ms | |
- | ERNIE-4.5-21B-A3B | 21B (3B active) | INT4 Q8 | 240 tok/s 198 tok/s | ~111 ms ~112 ms | |
- | ERNIE-4.5-21B-A3B-Base | 21B (3B active) | INT4 Q8 | 240 tok/s 198 tok/s | ~111 ms ~112 ms | |
- | ERNIE-4.5-VL-28B-A3B | 28B (3B active) | INT4 Q8 | 235 tok/s 186 tok/s | ~117 ms ~127 ms | |
- | ERNIE-4.5-VL-28B-A3B-Base | 28B (3B active) | INT4 Q8 | 235 tok/s 186 tok/s | ~117 ms ~127 ms | |
- | FrogNano 4B | 4B | INT4 Q8 FP16 | 234 tok/s 183 tok/s 124 tok/s | ~119 ms ~120 ms ~123 ms | |
- | Hy-Memory 30B A3B | 30B (3B active) | INT4 Q8 | 234 tok/s 175 tok/s | ~119 ms ~139 ms | |
- | Hy-MT2 30B A3B | 30B (3B active) | INT4 Q8 | 234 tok/s 175 tok/s | ~119 ms ~139 ms | |
- | Spark X2.5 4B | 4B | INT4 Q8 FP16 | 234 tok/s 183 tok/s 124 tok/s | ~119 ms ~120 ms ~123 ms | |
- | MiniCPM-V 4.7 35B A3B | 35B (3B active) | INT4 | 232 tok/s | ~121 ms | |
- | EmbeddingGemma 2 | 2B | INT4 Q8 FP16 | 229 tok/s 204 tok/s 161 tok/s | ~67 ms ~68 ms ~69 ms | |
- | Phi-2 | 2.7B | INT4 Q8 FP16 | 213 tok/s 185 tok/s 139 tok/s | ~86 ms ~86 ms ~88 ms | |
- | MaLLaM-3B | 3B | INT4 Q8 FP16 | 207 tok/s 177 tok/s 131 tok/s | ~94 ms ~94 ms ~96 ms | |
- | SmolLM3 3B | 3B | INT4 Q8 FP16 | 207 tok/s 177 tok/s 131 tok/s | ~94 ms ~94 ms ~96 ms | |
- | Falcon-7B | 7B | INT4 Q8 FP16 | 191 tok/s 136 tok/s 84 tok/s | ~198 ms ~200 ms ~205 ms | |
- | Falcon3-7B | 7B | INT4 Q8 FP16 | 186 tok/s 134 tok/s 83 tok/s | ~198 ms ~200 ms ~205 ms | |
- | Hy-MT2 7B | 7B | INT4 Q8 FP16 | 186 tok/s 134 tok/s 83 tok/s | ~198 ms ~200 ms ~205 ms | |
- | Mistral-7B-v0.1 | 7.3B | INT4 Q8 FP16 | 182 tok/s 130 tok/s 80 tok/s | ~206 ms ~208 ms ~213 ms | |
- | Youtu Parsing Omni | 8B | INT4 Q8 FP16 | 177 tok/s 124 tok/s 75 tok/s | ~224 ms ~227 ms ~232 ms | |
- | LensVLM 9B | 9B | INT4 Q8 FP16 | 163 tok/s 113 tok/s 68 tok/s | ~251 ms ~254 ms ~260 ms | |
- | MiMo V2.6 Distill Qwen 9B | 9B | INT4 Q8 FP16 | 163 tok/s 113 tok/s 68 tok/s | ~251 ms ~254 ms ~260 ms | |
- | ChatGLM3-6B-32K | 6B | INT4 Q8 FP16 | 160 tok/s 126 tok/s 84 tok/s | ~173 ms ~175 ms ~179 ms | |
- | Yi-6B | 6B | INT4 Q8 FP16 | 160 tok/s 126 tok/s 84 tok/s | ~173 ms ~175 ms ~179 ms | |
- | Falcon3-10B | 10B | INT4 Q8 FP16 | 154 tok/s 105 tok/s 62 tok/s | ~277 ms ~280 ms ~287 ms | |
- | Kimi-VL-A3B-Instruct | 16B (3B active) | INT4 Q8 | 154 tok/s 139 tok/s | ~110 ms ~110 ms | |
- | Falcon2-11B | 11B | INT4 Q8 FP16 | 150 tok/s 100 tok/s 58 tok/s | ~304 ms ~307 ms ~314 ms | |
- | Hunyuan Lite | 7B | INT4 Q8 FP16 | 149 tok/s 115 tok/s 75 tok/s | ~199 ms ~201 ms ~206 ms | |
- | OLMo 3 7B Base | 7B | INT4 Q8 FP16 | 149 tok/s 115 tok/s 75 tok/s | ~199 ms ~201 ms ~206 ms | |
- | SEA-LION-7B | 7.1B | INT4 Q8 FP16 | 148 tok/s 114 tok/s 74 tok/s | ~202 ms ~204 ms ~209 ms | |
- | SEA-LION-7B-Instruct | 7.1B | INT4 Q8 FP16 | 148 tok/s 114 tok/s 74 tok/s | ~202 ms ~204 ms ~209 ms | |
- | Sahabat-AI-Llama3-8B-Instruct | 8B | INT4 Q8 FP16 | 139 tok/s 106 tok/s 68 tok/s | ~226 ms ~228 ms ~233 ms | |
- | Typhoon-2-8B | 8B | INT4 Q8 FP16 | 139 tok/s 106 tok/s 68 tok/s | ~226 ms ~228 ms ~233 ms | |
- | GLM-4-9B | 9B | INT4 Q8 FP16 | 129 tok/s 97 tok/s 62 tok/s | ~253 ms ~255 ms ~261 ms | |
- | GLM-4-9B-Chat | 9B | INT4 Q8 FP16 | 129 tok/s 97 tok/s 62 tok/s | ~253 ms ~255 ms ~261 ms | |
- | GLM-4-9B-Chat-1M | 9B | INT4 Q8 FP16 | 129 tok/s 97 tok/s 62 tok/s | ~253 ms ~255 ms ~261 ms | |
- | Yi-9B | 9B | INT4 Q8 FP16 | 129 tok/s 97 tok/s 62 tok/s | ~253 ms ~255 ms ~261 ms | |
- | Sahabat-AI-Gemma2-9B | 9.2B | INT4 Q8 FP16 | 128 tok/s 96 tok/s 61 tok/s | ~258 ms ~260 ms ~266 ms | |
- | Sahabat-AI-Gemma2-9B-Instruct | 9.2B | INT4 Q8 FP16 | 128 tok/s 96 tok/s 61 tok/s | ~258 ms ~260 ms ~266 ms | |
- | GreenMind-14B-R1 | 14B | INT4 Q8 FP16 | 98 tok/s 70 tok/s 38 tok/s | ~386 ms ~390 ms ~466 ms | |
- | AREX-2 | 27B | INT4 Q8 | 78 tok/s 46 tok/s | ~723 ms ~792 ms | |
- | PPLX Decider v1 27B | 27B | INT4 Q8 | 78 tok/s 46 tok/s | ~723 ms ~792 ms | |
- | PPLX Decider v1.1 27B | 27B | INT4 Q8 | 78 tok/s 46 tok/s | ~723 ms ~792 ms | |
- | Agnes 3.0 Flash | 33B | INT4 | 67 tok/s | ~878 ms | |
- | Falcon-40B | 40B | INT4 | 58 tok/s | ~1.1s | |
Inference Capacity
Maximum model parameter class runnable by weight precision.
Context:
4-bit
32B Models
8-bit
14B Models
16-bit
14B Models
Fine-Tuning Capacity
Hardware capacity for local training and adapter fine-tuning.
Context Length:
QLoRA
32B Models
LoRA
14B Models
Full Parameter Training
3B Models
Workload Recommendations
Guidance for optimal precision and operational ceilings on NVIDIA RTX 5090 (32GB).
Inference Sweet Spot
Optimized for 14B models at uncompressed or Q8 precision, and 32B models at Q4 with up to 32k context.
Bandwidth & Speed Profile
With 1792 GB/s aggregate bandwidth, batch size 1 inference operates in a memory-bandwidth bound regime. Generates approximately 398 tok/s on an 8B Q4 model and 47 tok/s on a 70B Q4 model.
Frequently Asked Questions
©2025 ApX Machine Learning