AMD RX 7900 XT (20GB)
The AMD RX 7900 XT (20GB) provides 20 GB of memory with 800 GB/s bandwidth, supporting local inference via ROCm and Vulkan acceleration.
Memory
20 GB
Usable Memory Ceiling
20 GB
Bandwidth
800 GB/s
Max Inference Class
14B Models (Q4)
TDP
315W
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 AMD RX 7900 XT (20GB)
Inference throughput, memory compatibility, and generation speeds for open-weights LLMs on AMD RX 7900 XT (20GB).
Sort by:
Context:
KV Cache:
GPUs:
| Rank | Model | Parameters | Precision | TPS (Approx.) | TTFT | Can Run |
|---|---|---|---|---|---|---|
#48 | Qwen 3.8 27B | 27B | INT4 | 38 tok/s | ~3.3s | |
#83 | K2 Horizon MoVA 36B A4B | 36B (4B active) | INT4 | 114 tok/s | ~757 ms | |
#93 | Qwen3.5-27B | 27B | INT4 | 38 tok/s | ~3.3s | |
#105 | Sarvam-30B | 32B (2.4B active) | INT4 | 144 tok/s | ~490 ms | |
#106 | Qwen3.6 35B A3B | 35B (3B active) | INT4 | 127 tok/s | ~617 ms | |
#110 | Muse Glimmer 30B | 30B | INT4 | 33 tok/s | ~3.9s | |
#111 | MiMo V2 Flash | 15B (309B active) | INT4 | 4 tok/s | ~35.6s | |
#112 | Phi-4 Reasoning Plus | 14B | INT4 Q8 | 49 tok/s 32 tok/s | ~1.7s ~1.9s | |
#117 | Gemma 4 26B A4B | 25.2B (3.8B active) | INT4 | 133 tok/s | ~590 ms | |
#125 | Qwen3.5-9B | 9B | INT4 Q8 FP16 | 89 tok/s 58 tok/s 30 tok/s | ~1.1s ~1.1s ~1.3s | |
#126 | Qwen3.5-35B-A3B | 35B (3B active) | INT4 | 127 tok/s | ~617 ms | |
#128 | K2 Horizon 7B | 7B | INT4 Q8 FP16 | 105 tok/s 70 tok/s 41 tok/s | ~875 ms ~880 ms ~890 ms | |
#135 | Gemma 4 12B | 11.95B | INT4 Q8 | 55 tok/s 37 tok/s | ~1.5s ~1.6s | |
#140 | Qwen3.5-4B | 4B | INT4 Q8 FP16 | 142 tok/s 103 tok/s 64 tok/s | ~516 ms ~519 ms ~525 ms | |
#143 | GPT-OSS 20B | 21B (3.6B active) | INT4 | 70 tok/s | ~606 ms | |
#144 | GLM-4V | 9B | INT4 Q8 | 68 tok/s 49 tok/s | ~1.1s ~1.1s | |
#150 | NVIDIA Nemotron 3 Nano 30B-A3B | 3.5B (30B active) | INT4 Q8 FP16 | 33 tok/s 20 tok/s 11 tok/s | ~3.5s ~3.5s ~3.6s | |
#153 | Qwen3-30B-A3B | 30B (3B active) | INT4 | 135 tok/s | ~559 ms | |
#158 | Granite 4.2 8B | 8B | INT4 Q8 FP16 | 97 tok/s 64 tok/s 35 tok/s | ~995 ms ~1.0s ~1.1s | |
#159 | Gemma 4 E4B | 8B | INT4 Q8 FP16 | 97 tok/s 64 tok/s 35 tok/s | ~995 ms ~1.0s ~1.1s | |
#166 | Phi-4 | 14B | INT4 Q8 | 65 tok/s 40 tok/s | ~1.7s ~1.7s | |
#167 | Gemma 3 27B | 27B | INT4 | 38 tok/s | ~3.3s | |
#170 | DeepSeek-R1 32B | 32B | INT4 | 32 tok/s | ~4.2s | |
#171 | Qwen3-32B | 32B | INT4 | 31 tok/s | ~4.2s | |
#175 | Magistral Small | 24B | INT4 | 31 tok/s | ~3.2s | |
#177 | Qwen2.5-32B | 32B | INT4 | 31 tok/s | ~4.2s | |
#178 | OLMo 3 7B Think | 7B | INT4 Q8 FP16 | 80 tok/s 59 tok/s 35 tok/s | ~878 ms ~883 ms ~969 ms | |
#179 | DeepSeek-R1 14B | 14B | INT4 Q8 | 67 tok/s 41 tok/s | ~1.7s ~1.7s | |
#183 | Ministral 3 14B | 14B | INT4 Q8 | 49 tok/s 32 tok/s | ~1.7s ~1.9s | |
#186 | Gemma 3 12B | 12B | INT4 Q8 | 73 tok/s 46 tok/s | ~1.5s ~1.5s | |
#188 | Mistral-Small-2501 | 24B | INT4 | 42 tok/s | ~2.9s | |
#189 | Ministral 3 8B | 8B | INT4 Q8 FP16 | 74 tok/s 54 tok/s 30 tok/s | ~998 ms ~1.0s ~1.2s | |
#190 | Gemma 4 31B | 30.7B | INT4 | 33 tok/s | ~4.0s | |
#191 | MiniCPM5-2B | 2B | INT4 Q8 FP16 | 186 tok/s 150 tok/s 104 tok/s | ~278 ms ~279 ms ~282 ms | |
#192 | Qwen3.5-2B | 2B | INT4 Q8 FP16 | 186 tok/s 150 tok/s 104 tok/s | ~278 ms ~279 ms ~282 ms | |
#193 | Gemma 2 27B | 27B | INT4 | 38 tok/s | ~3.3s | |
#194 | Qwen2.5-14B | 14B | INT4 Q8 | 65 tok/s 40 tok/s | ~1.7s ~1.7s | |
#200 | Gemma 4 E2B | 5.1B | INT4 Q8 FP16 | 126 tok/s 88 tok/s 53 tok/s | ~648 ms ~651 ms ~659 ms | |
#202 | DeepSeek-R1 8B | 8B | INT4 Q8 FP16 | 99 tok/s 65 tok/s 35 tok/s | ~995 ms ~1.0s ~1.1s | |
#204 | Qwen3-4B | 4B | INT4 Q8 FP16 | 142 tok/s 103 tok/s 64 tok/s | ~516 ms ~519 ms ~525 ms | |
#205 | Ministral 3 3B | 3B | INT4 Q8 FP16 | 121 tok/s 99 tok/s 69 tok/s | ~399 ms ~400 ms ~405 ms | |
#207 | Gemma 2 9B | 9B | INT4 Q8 FP16 | 89 tok/s 58 tok/s 30 tok/s | ~1.1s ~1.1s ~1.3s | |
#208 | Qwen3.5-0.8B | 0.8B | INT4 Q8 FP16 | 235 tok/s 209 tok/s 168 tok/s | ~118 ms ~118 ms ~120 ms | |
#209 | Qwen3-14B | 14B | INT4 Q8 | 65 tok/s 40 tok/s | ~1.7s ~1.7s | |
#210 | Gemma 3 4B | 4B | INT4 Q8 FP16 | 142 tok/s 103 tok/s 64 tok/s | ~516 ms ~519 ms ~525 ms | |
#211 | Qwen3-8B | 8B | INT4 Q8 FP16 | 97 tok/s 64 tok/s 35 tok/s | ~995 ms ~1.0s ~1.1s | |
#213 | DeepSeek-R1 7B | 7B | INT4 Q8 FP16 | 107 tok/s 71 tok/s 41 tok/s | ~875 ms ~880 ms ~890 ms | |
#214 | Qwen3-1.7B | 1.7B | INT4 Q8 FP16 | 196 tok/s 161 tok/s 115 tok/s | ~239 ms ~240 ms ~242 ms | |
#215 | Llama 3.1 8B | 8B | INT4 Q8 FP16 | 97 tok/s 64 tok/s 35 tok/s | ~995 ms ~1.0s ~1.1s | |
#216 | Gemma 3 270M | 0.27B | INT4 Q8 FP16 | 254 tok/s 244 tok/s 222 tok/s | ~45 ms ~45 ms ~45 ms | |
#218 | Qwen3-0.6B | 0.6B | INT4 Q8 FP16 | 247 tok/s 225 tok/s 187 tok/s | ~89 ms ~90 ms ~90 ms | |
#219 | Llama 3 8B | 8B | INT4 Q8 FP16 | 97 tok/s 64 tok/s 35 tok/s | ~995 ms ~1.0s ~1.1s | |
#221 | Phi-4-Mini | 3.8B | INT4 Q8 FP16 | 146 tok/s 107 tok/s 67 tok/s | ~492 ms ~495 ms ~500 ms | |
#223 | Qwen2.5-7B | 7B | INT4 Q8 FP16 | 105 tok/s 70 tok/s 41 tok/s | ~875 ms ~880 ms ~890 ms | |
#224 | Ministral-8B-2410 | 8B | INT4 Q8 FP16 | 97 tok/s 64 tok/s 35 tok/s | ~995 ms ~1.0s ~1.1s | |
#225 | DeepSeek-R1 1.5B | 1.5B | INT4 Q8 FP16 | 204 tok/s 169 tok/s 123 tok/s | ~212 ms ~213 ms ~215 ms | |
#226 | Llama 3.2 3B | 3B | INT4 Q8 FP16 | 161 tok/s 122 tok/s 79 tok/s | ~396 ms ~398 ms ~403 ms | |
#227 | Nemotron 3.5 Lightning 30B A3B | 30B (3B active) | INT4 | 135 tok/s | ~559 ms | |
#228 | Phi-3-mini | 3.8B | INT4 Q8 FP16 | 146 tok/s 107 tok/s 67 tok/s | ~492 ms ~495 ms ~500 ms | |
#229 | Gemma 3n E2B IT | 6B (2B active) | INT4 Q8 FP16 | 121 tok/s 107 tok/s 81 tok/s | ~299 ms ~300 ms ~303 ms | |
#230 | Mistral-7B-Instruct-v0.1 | 7.3B | INT4 Q8 FP16 | 102 tok/s 68 tok/s 39 tok/s | ~911 ms ~915 ms ~926 ms | |
#231 | Llama 3.2 1B | 1B | INT4 Q8 FP16 | 224 tok/s 195 tok/s 152 tok/s | ~147 ms ~148 ms ~150 ms | |
#232 | Gemma 2 2B | 2B | INT4 Q8 FP16 | 186 tok/s 150 tok/s 104 tok/s | ~278 ms ~279 ms ~282 ms | |
#233 | Phi-3-medium | 14B | INT4 Q8 | 65 tok/s 40 tok/s | ~1.7s ~1.7s | |
#234 | OLMo 3 7B Instruct | 7B | INT4 Q8 FP16 | 80 tok/s 59 tok/s 35 tok/s | ~878 ms ~883 ms ~969 ms | |
#236 | Qwen2-7B | 7B | INT4 Q8 FP16 | 105 tok/s 70 tok/s 41 tok/s | ~875 ms ~880 ms ~890 ms | |
#237 | Phi-3-small | 7B | INT4 Q8 FP16 | 105 tok/s 70 tok/s 41 tok/s | ~875 ms ~880 ms ~890 ms | |
#238 | Gemma 3 1B | 1B | INT4 Q8 FP16 | 224 tok/s 195 tok/s 152 tok/s | ~147 ms ~148 ms ~150 ms | |
#239 | Mistral-7B-Instruct-v0.2 | 7.3B | INT4 Q8 FP16 | 102 tok/s 68 tok/s 39 tok/s | ~911 ms ~915 ms ~926 ms | |
#241 | Gemma 1 7B | 7B | INT4 Q8 FP16 | 80 tok/s 59 tok/s 35 tok/s | ~878 ms ~883 ms ~969 ms | |
#242 | Gemma 1 2B | 2B | INT4 Q8 FP16 | 194 tok/s 154 tok/s 106 tok/s | ~278 ms ~279 ms ~282 ms | |
#243 | ChatGLM3-6B | 6B | INT4 Q8 FP16 | 87 tok/s 66 tok/s 42 tok/s | ~758 ms ~762 ms ~771 ms | |
#244 | ChatGLM2-6B | 6B | INT4 Q8 FP16 | 87 tok/s 66 tok/s 42 tok/s | ~758 ms ~762 ms ~771 ms | |
#245 | ChatGLM-6B | 6B | INT4 Q8 FP16 | 87 tok/s 66 tok/s 42 tok/s | ~758 ms ~762 ms ~771 ms | |
Unranked Models (Sorted by Speed) | ||||||
- | DINOv3 ViT-L/16 | 0.3B | INT4 Q8 FP16 | 265 tok/s 252 tok/s 227 tok/s | ~49 ms ~49 ms ~49 ms | |
- | Qwen2-0.5B | 0.5B | INT4 Q8 FP16 | 253 tok/s 233 tok/s 199 tok/s | ~75 ms ~76 ms ~76 ms | |
- | Qwen2.5-0.5B | 0.5B | INT4 Q8 FP16 | 253 tok/s 233 tok/s 199 tok/s | ~75 ms ~76 ms ~76 ms | |
- | ERNIE-4.5-0.3B | 0.3B | INT4 Q8 FP16 | 250 tok/s 239 tok/s 217 tok/s | ~49 ms ~49 ms ~49 ms | |
- | ERNIE-4.5-0.3B-Base | 0.3B | INT4 Q8 FP16 | 250 tok/s 239 tok/s 217 tok/s | ~49 ms ~49 ms ~49 ms | |
- | Liquid D1-Omni 600M | 0.6B | INT4 Q8 FP16 | 247 tok/s 225 tok/s 187 tok/s | ~89 ms ~90 ms ~90 ms | |
- | PixArt-α XL 2 1024-MS | 0.61B | INT4 Q8 FP16 | 246 tok/s 224 tok/s 186 tok/s | ~91 ms ~91 ms ~92 ms | |
- | PixArt-Σ XL 2 1024-MS | 0.61B | INT4 Q8 FP16 | 246 tok/s 224 tok/s 186 tok/s | ~91 ms ~91 ms ~92 ms | |
- | Nemotron 3.5 Parse | 0.63B | INT4 Q8 FP16 | 245 tok/s 222 tok/s 184 tok/s | ~94 ms ~94 ms ~95 ms | |
- | SD 2.1 Turbo | 0.86B | INT4 Q8 FP16 | 232 tok/s 205 tok/s 163 tok/s | ~126 ms ~127 ms ~128 ms | |
- | Stable Diffusion 1.4 | 0.86B | INT4 Q8 FP16 | 232 tok/s 205 tok/s 163 tok/s | ~126 ms ~127 ms ~128 ms | |
- | Stable Diffusion 2.1 | 0.86B | INT4 Q8 FP16 | 232 tok/s 205 tok/s 163 tok/s | ~126 ms ~127 ms ~128 ms | |
- | Falcon-1B | 1B | INT4 Q8 FP16 | 231 tok/s 200 tok/s 154 tok/s | ~147 ms ~148 ms ~149 ms | |
- | Falcon3-1B | 1B | INT4 Q8 FP16 | 224 tok/s 195 tok/s 152 tok/s | ~147 ms ~148 ms ~150 ms | |
- | MiniCPM-V 4.7 1B | 1B | INT4 Q8 FP16 | 224 tok/s 195 tok/s 152 tok/s | ~147 ms ~148 ms ~150 ms | |
- | Kandinsky 2.2 Decoder | 1.22B | INT4 Q8 FP16 | 215 tok/s 183 tok/s 138 tok/s | ~176 ms ~177 ms ~179 ms | |
- | Wan 2.1 T2V 1.3B | 1.3B | INT4 Q8 FP16 | 212 tok/s 179 tok/s 133 tok/s | ~186 ms ~187 ms ~189 ms | |
- | Qwen2-1.5B | 1.5B | INT4 Q8 FP16 | 204 tok/s 169 tok/s 123 tok/s | ~212 ms ~213 ms ~215 ms | |
- | Qwen2.5-1.5B | 1.5B | INT4 Q8 FP16 | 204 tok/s 169 tok/s 123 tok/s | ~212 ms ~213 ms ~215 ms | |
- | SVD Image-to-Video XT | 1.5B | INT4 Q8 FP16 | 204 tok/s 169 tok/s 123 tok/s | ~212 ms ~213 ms ~215 ms | |
- | SVD Image-to-Video XT 1.1 | 1.5B | INT4 Q8 FP16 | 204 tok/s 169 tok/s 123 tok/s | ~212 ms ~213 ms ~215 ms | |
- | LFM3 1.7B | 1.7B | INT4 Q8 FP16 | 196 tok/s 161 tok/s 115 tok/s | ~239 ms ~240 ms ~242 ms | |
- | LFM3 1.7B Base | 1.7B | INT4 Q8 FP16 | 196 tok/s 161 tok/s 115 tok/s | ~239 ms ~240 ms ~242 ms | |
- | Spark X2.5 1.7B | 1.7B | INT4 Q8 FP16 | 196 tok/s 161 tok/s 115 tok/s | ~239 ms ~240 ms ~242 ms | |
- | Hy-MT2 1.8B | 1.8B | INT4 Q8 FP16 | 193 tok/s 157 tok/s 111 tok/s | ~251 ms ~252 ms ~255 ms | |
- | CogVideoX 2B | 2B | INT4 Q8 FP16 | 186 tok/s 150 tok/s 104 tok/s | ~278 ms ~279 ms ~282 ms | |
- | LFM3 Audio 2B | 2B | INT4 Q8 FP16 | 186 tok/s 150 tok/s 104 tok/s | ~278 ms ~279 ms ~282 ms | |
- | LFM3 Omni 2B | 2B | INT4 Q8 FP16 | 186 tok/s 150 tok/s 104 tok/s | ~278 ms ~279 ms ~282 ms | |
- | LFM3 VL 2B | 2B | INT4 Q8 FP16 | 186 tok/s 150 tok/s 104 tok/s | ~278 ms ~279 ms ~282 ms | |
- | Liquid D1 2B | 2B | INT4 Q8 FP16 | 186 tok/s 150 tok/s 104 tok/s | ~278 ms ~279 ms ~282 ms | |
- | LTX-Video 0.9.1 | 2B | INT4 Q8 FP16 | 186 tok/s 150 tok/s 104 tok/s | ~278 ms ~279 ms ~282 ms | |
- | Moondream 3.1 | 9B (2B active) | INT4 Q8 FP16 | 186 tok/s 150 tok/s 94 tok/s | ~312 ms ~313 ms ~370 ms | |
- | Stable Diffusion 3 Medium | 2B | INT4 Q8 FP16 | 186 tok/s 150 tok/s 104 tok/s | ~278 ms ~279 ms ~282 ms | |
- | Kandinsky 3.0 | 2.1B | INT4 Q8 FP16 | 184 tok/s 146 tok/s 101 tok/s | ~290 ms ~291 ms ~294 ms | |
- | LTX-2 | 2.2B | INT4 Q8 FP16 | 181 tok/s 143 tok/s 98 tok/s | ~301 ms ~303 ms ~306 ms | |
- | SDXL 1.0 Refiner | 2.31B | INT4 Q8 FP16 | 178 tok/s 140 tok/s 95 tok/s | ~315 ms ~316 ms ~319 ms | |
- | Stable Diffusion 3.5 Medium | 2.5B | INT4 Q8 FP16 | 173 tok/s 134 tok/s 90 tok/s | ~337 ms ~339 ms ~342 ms | |
- | SDXL 1.0 Base | 2.56B | INT4 Q8 FP16 | 172 tok/s 133 tok/s 89 tok/s | ~344 ms ~346 ms ~350 ms | |
- | SDXL Turbo | 2.56B | INT4 Q8 FP16 | 172 tok/s 133 tok/s 89 tok/s | ~344 ms ~346 ms ~350 ms | |
- | Falcon-3B | 3B | INT4 Q8 FP16 | 168 tok/s 125 tok/s 81 tok/s | ~396 ms ~398 ms ~403 ms | |
- | CroissantLLM Base | 1.3B | INT4 Q8 FP16 | 166 tok/s 147 tok/s 115 tok/s | ~187 ms ~188 ms ~190 ms | |
- | Phi-1 | 1.3B | INT4 Q8 FP16 | 166 tok/s 147 tok/s 115 tok/s | ~187 ms ~188 ms ~190 ms | |
- | Phi-1.5 | 1.3B | INT4 Q8 FP16 | 166 tok/s 147 tok/s 115 tok/s | ~187 ms ~188 ms ~190 ms | |
- | DeepSeek-R1 3B | 3B | INT4 Q8 FP16 | 163 tok/s 123 tok/s 80 tok/s | ~396 ms ~398 ms ~403 ms | |
- | Mellum 2.1 12B A2.5B Thinking | 12B (2.5B active) | INT4 Q8 | 163 tok/s 129 tok/s | ~387 ms ~389 ms | |
- | Falcon3-3B | 3B | INT4 Q8 FP16 | 161 tok/s 122 tok/s 79 tok/s | ~396 ms ~398 ms ~403 ms | |
- | LFM2-VL 3B | 3B | INT4 Q8 FP16 | 161 tok/s 122 tok/s 79 tok/s | ~396 ms ~398 ms ~403 ms | |
- | Liquid D1 3B | 3B | INT4 Q8 FP16 | 161 tok/s 122 tok/s 79 tok/s | ~396 ms ~398 ms ~403 ms | |
- | Liquid D1-VL 3B | 3B | INT4 Q8 FP16 | 161 tok/s 122 tok/s 79 tok/s | ~396 ms ~398 ms ~403 ms | |
- | Liquid LFM2-VL 3B D1 | 3B | INT4 Q8 FP16 | 161 tok/s 122 tok/s 79 tok/s | ~396 ms ~398 ms ~403 ms | |
- | Qwen2.5-3B | 3B | INT4 Q8 FP16 | 161 tok/s 122 tok/s 79 tok/s | ~396 ms ~398 ms ~403 ms | |
- | LFM2.5 24B A2B | 24B (2.5B active) | INT4 | 154 tok/s | ~438 ms | |
- | ERNIE-4.5-21B-A3B | 21B (3B active) | INT4 | 147 tok/s | ~483 ms | |
- | ERNIE-4.5-21B-A3B-Base | 21B (3B active) | INT4 | 147 tok/s | ~483 ms | |
- | FLUX.2 Klein 4B | 4B | INT4 Q8 FP16 | 142 tok/s 103 tok/s 64 tok/s | ~516 ms ~519 ms ~525 ms | |
- | FrogNano 4B | 4B | INT4 Q8 FP16 | 142 tok/s 103 tok/s 64 tok/s | ~516 ms ~519 ms ~525 ms | |
- | Mage-VL Cowork | 4B | INT4 Q8 FP16 | 142 tok/s 103 tok/s 64 tok/s | ~516 ms ~519 ms ~525 ms | |
- | Spark X2.5 4B | 4B | INT4 Q8 FP16 | 142 tok/s 103 tok/s 64 tok/s | ~516 ms ~519 ms ~525 ms | |
- | EmbeddingGemma 2 | 2B | INT4 Q8 FP16 | 138 tok/s 119 tok/s 88 tok/s | ~280 ms ~281 ms ~284 ms | |
- | ERNIE-4.5-VL-28B-A3B | 28B (3B active) | INT4 | 136 tok/s | ~551 ms | |
- | ERNIE-4.5-VL-28B-A3B-Base | 28B (3B active) | INT4 | 136 tok/s | ~551 ms | |
- | Hy-Memory 30B A3B | 30B (3B active) | INT4 | 135 tok/s | ~559 ms | |
- | Hy-MT2 30B A3B | 30B (3B active) | INT4 | 135 tok/s | ~559 ms | |
- | CogVideoX 1.5 5B | 5B | INT4 Q8 FP16 | 127 tok/s 89 tok/s 54 tok/s | ~636 ms ~639 ms ~647 ms | |
- | CogVideoX 5B | 5B | INT4 Q8 FP16 | 127 tok/s 89 tok/s 54 tok/s | ~636 ms ~639 ms ~647 ms | |
- | MiniCPM-V 4.7 35B A3B | 35B (3B active) | INT4 | 127 tok/s | ~617 ms | |
- | Wan 2.2 TI2V 5B | 5B | INT4 Q8 FP16 | 127 tok/s 89 tok/s 54 tok/s | ~636 ms ~639 ms ~647 ms | |
- | Phi-2 | 2.7B | INT4 Q8 FP16 | 126 tok/s 104 tok/s 74 tok/s | ~363 ms ~365 ms ~369 ms | |
- | MaLLaM-3B | 3B | INT4 Q8 FP16 | 121 tok/s 99 tok/s 69 tok/s | ~399 ms ~400 ms ~405 ms | |
- | SmolLM3 3B | 3B | INT4 Q8 FP16 | 121 tok/s 99 tok/s 69 tok/s | ~399 ms ~400 ms ~405 ms | |
- | Falcon-7B | 7B | INT4 Q8 FP16 | 109 tok/s 72 tok/s 41 tok/s | ~875 ms ~880 ms ~890 ms | |
- | AuraFlow v0.2 | 6.8B | INT4 Q8 FP16 | 107 tok/s 72 tok/s 42 tok/s | ~851 ms ~856 ms ~866 ms | |
- | AuraFlow v0.3 | 6.8B | INT4 Q8 FP16 | 107 tok/s 72 tok/s 42 tok/s | ~851 ms ~856 ms ~866 ms | |
- | Falcon3-7B | 7B | INT4 Q8 FP16 | 105 tok/s 70 tok/s 41 tok/s | ~875 ms ~880 ms ~890 ms | |
- | Hy-MT2 7B | 7B | INT4 Q8 FP16 | 105 tok/s 70 tok/s 41 tok/s | ~875 ms ~880 ms ~890 ms | |
- | Mistral-7B-v0.1 | 7.3B | INT4 Q8 FP16 | 102 tok/s 68 tok/s 39 tok/s | ~911 ms ~915 ms ~926 ms | |
- | Youtu Parsing Omni | 8B | INT4 Q8 FP16 | 99 tok/s 65 tok/s 35 tok/s | ~995 ms ~1.0s ~1.1s | |
- | Stable Diffusion 3.5 Large | 8.1B | INT4 Q8 FP16 | 96 tok/s 63 tok/s 34 tok/s | ~1.0s ~1.0s ~1.1s | |
- | Stable Diffusion 3.5 Large Turbo | 8.1B | INT4 Q8 FP16 | 96 tok/s 63 tok/s 34 tok/s | ~1.0s ~1.0s ~1.1s | |
- | FLUX.2 Klein 9B | 9B | INT4 Q8 FP16 | 89 tok/s 58 tok/s 30 tok/s | ~1.1s ~1.1s ~1.3s | |
- | LensVLM 9B | 9B | INT4 Q8 FP16 | 89 tok/s 58 tok/s 30 tok/s | ~1.1s ~1.1s ~1.3s | |
- | MiMo V2.6 Distill Qwen 9B | 9B | INT4 Q8 FP16 | 89 tok/s 58 tok/s 30 tok/s | ~1.1s ~1.1s ~1.3s | |
- | ChatGLM3-6B-32K | 6B | INT4 Q8 FP16 | 87 tok/s 66 tok/s 42 tok/s | ~758 ms ~762 ms ~771 ms | |
- | Yi-6B | 6B | INT4 Q8 FP16 | 87 tok/s 66 tok/s 42 tok/s | ~758 ms ~762 ms ~771 ms | |
- | Falcon3-10B | 10B | INT4 Q8 | 83 tok/s 53 tok/s | ~1.2s ~1.2s | |
- | Kimi-VL-A3B-Instruct | 16B (3B active) | INT4 Q8 | 83 tok/s 66 tok/s | ~469 ms ~551 ms | |
- | Mochi 1 Preview | 10B | INT4 Q8 | 83 tok/s 53 tok/s | ~1.2s ~1.2s | |
- | Falcon2-11B | 11B | INT4 Q8 | 80 tok/s 50 tok/s | ~1.4s ~1.4s | |
- | Hunyuan Lite | 7B | INT4 Q8 FP16 | 80 tok/s 59 tok/s 35 tok/s | ~878 ms ~883 ms ~969 ms | |
- | OLMo 3 7B Base | 7B | INT4 Q8 FP16 | 80 tok/s 59 tok/s 35 tok/s | ~878 ms ~883 ms ~969 ms | |
- | SEA-LION-7B | 7.1B | INT4 Q8 FP16 | 79 tok/s 58 tok/s 34 tok/s | ~890 ms ~895 ms ~982 ms | |
- | SEA-LION-7B-Instruct | 7.1B | INT4 Q8 FP16 | 79 tok/s 58 tok/s 34 tok/s | ~890 ms ~895 ms ~982 ms | |
- | Sahabat-AI-Llama3-8B-Instruct | 8B | INT4 Q8 FP16 | 74 tok/s 54 tok/s 30 tok/s | ~998 ms ~1.0s ~1.2s | |
- | Typhoon-2-8B | 8B | INT4 Q8 FP16 | 74 tok/s 54 tok/s 30 tok/s | ~998 ms ~1.0s ~1.2s | |
- | FLUX.1 Canny Dev | 11.9B | INT4 Q8 | 73 tok/s 46 tok/s | ~1.5s ~1.5s | |
- | FLUX.1 Depth Dev | 11.9B | INT4 Q8 | 73 tok/s 46 tok/s | ~1.5s ~1.5s | |
- | FLUX.1 Dev | 11.9B | INT4 Q8 | 73 tok/s 46 tok/s | ~1.5s ~1.5s | |
- | FLUX.1 Fill Dev | 11.9B | INT4 Q8 | 73 tok/s 46 tok/s | ~1.5s ~1.5s | |
- | FLUX.1 Schnell | 11.9B | INT4 Q8 | 73 tok/s 46 tok/s | ~1.5s ~1.5s | |
- | HunyuanVideo 1.5 720P T2V | 13B | INT4 Q8 | 69 tok/s 43 tok/s | ~1.6s ~1.6s | |
- | GLM-4-9B | 9B | INT4 Q8 | 68 tok/s 49 tok/s | ~1.1s ~1.1s | |
- | GLM-4-9B-Chat | 9B | INT4 Q8 | 68 tok/s 49 tok/s | ~1.1s ~1.1s | |
- | GLM-4-9B-Chat-1M | 9B | INT4 Q8 | 68 tok/s 49 tok/s | ~1.1s ~1.1s | |
- | Yi-9B | 9B | INT4 Q8 | 68 tok/s 49 tok/s | ~1.1s ~1.1s | |
- | Sahabat-AI-Gemma2-9B | 9.2B | INT4 Q8 | 67 tok/s 48 tok/s | ~1.1s ~1.1s | |
- | Sahabat-AI-Gemma2-9B-Instruct | 9.2B | INT4 Q8 | 67 tok/s 48 tok/s | ~1.1s ~1.1s | |
- | Wan 2.1 I2V 14B 720P | 14B | INT4 Q8 | 65 tok/s 40 tok/s | ~1.7s ~1.7s | |
- | Wan 2.1 T2V 14B | 14B | INT4 Q8 | 65 tok/s 40 tok/s | ~1.7s ~1.7s | |
- | Wan 2.2 T2V A14B | 14B | INT4 Q8 | 65 tok/s 40 tok/s | ~1.7s ~1.7s | |
- | FLUX.2 Dev | 15B | INT4 Q8 | 62 tok/s 36 tok/s | ~1.8s ~2.0s | |
- | GreenMind-14B-R1 | 14B | INT4 Q8 | 49 tok/s 32 tok/s | ~1.7s ~1.9s | |
- | Arete 27B | 27B | INT4 | 38 tok/s | ~3.3s | |
- | AREX-2 | 27B | INT4 | 38 tok/s | ~3.3s | |
- | PPLX Decider v1 27B | 27B | INT4 | 38 tok/s | ~3.3s | |
- | PPLX Decider v1.1 27B | 27B | INT4 | 38 tok/s | ~3.3s | |
- | Agnes 3.0 Flash | 33B | INT4 | 31 tok/s | ~4.3s | |
Inference Capacity
Maximum model parameter class runnable by weight precision.
Context:
4-bit
14B Models
8-bit
14B Models
16-bit
8B Models
Fine-Tuning Capacity
Hardware capacity for local training and adapter fine-tuning.
Context Length:
QLoRA
32B Models
LoRA
8B Models
Full Parameter Training
1B Models
Workload Recommendations
Guidance for optimal precision and operational ceilings on AMD RX 7900 XT (20GB).
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 800 GB/s aggregate bandwidth, batch size 1 inference operates in a memory-bandwidth bound regime. Generates approximately 178 tok/s on an 8B Q4 model and 21 tok/s on a 70B Q4 model.
Frequently Asked Questions
©2025 ApX Machine Learning