AMD · unified x86

AMD Ryzen AI Max 385 (Radeon 8050S)

AMD Ryzen AI Max 385 (Radeon 8050S) has 32 GB of VRAM at 256 GB/s — about 22.32 GiB usable after driver and compositor overhead. 1985 of 2118 indexed models fit at 4K context with f16 KV. Note only 24 GB of its 32 GB is allocatable to the GPU.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
32 GB
LPDDR5X-8000
Bandwidth
256 GB/s
256-bit bus
Tensor FP16
—
dense
TDP
120 W
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1705vision language 176audio tts 21image 2video 16audio asr 39embedding 26

What fits at 4K context

largest quantization that fits, per model · 1985 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Qwen3-Coder-Next-REAMMoEI1-IQ3_XS60.3B23.19 GiB0.09 GiB23.98 GiB0.02 GiB37±37%
WizardLM-Uncensored-SuperCOT-StoryTelling-30bQ4_032.5B17.10 GiB6.09 GiB23.96 GiB0.04 GiB6±25%
Wizard-Vicuna-30B-UncensoredQ4_032.5B17.10 GiB6.09 GiB23.96 GiB0.04 GiB6±25%
Qwen3.5-122B-A10B-hereticMoEI1-IQ1_S123B23.13 GiB0.09 GiB23.95 GiB0.05 GiB35±37%
Gemma4-Gutenberg-31BQ5_K_L31.3B21.37 GiB1.80 GiB23.95 GiB0.05 GiB6±25%
gemma-4-31B-itQ5_K_L31.3B21.37 GiB1.80 GiB23.95 GiB0.05 GiB6±25%
Gemma4-Gutenberg-31B-HereticQ5_K_L31.3B21.37 GiB1.80 GiB23.95 GiB0.05 GiB6±25%
Equinox-31BQ5_K_L31.3B21.37 GiB1.80 GiB23.95 GiB0.05 GiB6±25%
gemma-4-31B-it-SDFT-Heretic-RPQ5_K_L30.7B21.37 GiB1.80 GiB23.95 GiB0.05 GiB6±25%
solar-pro-preview-instructKV unresolvedQ8_022.1B21.91 GiB1.25 GiB23.92 GiB0.08 GiB6±25%
Llama-3_1-Nemotron-51B-InstructIQ2_XXS51.5B13.07 GiB10.00 GiB23.91 GiB0.09 GiB6±25%
GLM-4.7-FlashMoEQ6_K31.2B23.00 GiB0.21 GiB23.91 GiB0.09 GiB26±37%
Nemotron-Cascade-2-30B-A3BMoEQ4_K_M31.6B23.03 GiB0.20 GiB23.91 GiB0.09 GiB30±37%
CallerQ5_K_L32.8B22.11 GiB1.00 GiB23.91 GiB0.09 GiB6±25%
Dumpling-Qwen2.5-32BQ5_K_L32.8B22.11 GiB1.00 GiB23.91 GiB0.09 GiB6±25%
OREAL-32BQ5_K_L32.8B22.11 GiB1.00 GiB23.91 GiB0.09 GiB6±25%
openhands-lm-32b-v0.1Q5_K_L32.8B22.11 GiB1.00 GiB23.91 GiB0.09 GiB6±25%
LongWriter-Zero-32BQ5_K_L32.8B22.11 GiB1.00 GiB23.91 GiB0.09 GiB6±25%
OpenCodeReasoning-Nemotron-32B-IOIQ5_K_L32.8B22.11 GiB1.00 GiB23.91 GiB0.09 GiB6±25%
Qwen2.5-Coder-32B-Instruct-abliteratedQ5_K_L32.8B22.11 GiB1.00 GiB23.91 GiB0.09 GiB6±25%
OlympicCoder-32BQ5_K_L32.8B22.11 GiB1.00 GiB23.91 GiB0.09 GiB6±25%
OpenCodeReasoning-Nemotron-32BQ5_K_L32.8B22.11 GiB1.00 GiB23.91 GiB0.09 GiB6±25%
OpenThinker-32BQ5_K_L32.8B22.11 GiB1.00 GiB23.91 GiB0.09 GiB6±25%
QwQ-32B-ArliAI-RpR-v4Q5_K_L32.8B22.11 GiB1.00 GiB23.91 GiB0.09 GiB6±25%
Qwen2.5-Coder-32B-InstructQ5_K_L32.8B22.11 GiB1.00 GiB23.91 GiB0.09 GiB6±25%
QwQ-32B-abliteratedQ5_K_L32.8B22.11 GiB1.00 GiB23.91 GiB0.09 GiB6±25%
OpenThinker2-32BQ5_K_L32.8B22.11 GiB1.00 GiB23.91 GiB0.09 GiB6±25%
INTELLECT-2Q5_K_L32.8B22.11 GiB1.00 GiB23.91 GiB0.09 GiB6±25%
Qwen2.5-32B-InstructQ5_K_L32.8B22.11 GiB1.00 GiB23.91 GiB0.09 GiB6±25%
QwQ-32B-PreviewQ5_K_L32.8B22.11 GiB1.00 GiB23.91 GiB0.09 GiB6±25%
Qwen2.5-Coder-32BQ5_K_L32.8B22.11 GiB1.00 GiB23.91 GiB0.09 GiB6±25%
Qwen2.5-32b-RP-InkQ5_K_L32.8B22.11 GiB1.00 GiB23.91 GiB0.09 GiB6±25%
TinyR1-32B-PreviewQ5_K_L32.8B22.11 GiB1.00 GiB23.91 GiB0.09 GiB6±25%
deepseek-r1-qwen-2.5-32B-ablatedQ5_K_L32.8B22.11 GiB1.00 GiB23.91 GiB0.09 GiB6±25%
Rombos-LLM-V2.5-Qwen-32bQ5_K_L32.8B22.11 GiB1.00 GiB23.91 GiB0.09 GiB6±25%
DeepSeek-R1-Distill-Qwen-32B-abliteratedQ5_K_L32.8B22.11 GiB1.00 GiB23.91 GiB0.09 GiB6±25%
Qwen2.5-32B-ArliAI-RPMax-v1.3Q5_K_L32.8B22.11 GiB1.00 GiB23.91 GiB0.09 GiB6±25%
DeepSeek-R1-Distill-Qwen-32BQ5_K_L32.8B22.11 GiB1.00 GiB23.91 GiB0.09 GiB6±25%
Qwen2.5-VL-32B-InstructQ5_K_L33.5B22.11 GiB1.00 GiB23.91 GiB0.09 GiB6±25%
EVA-Qwen2.5-32B-v0.2Q5_K_L32.8B22.11 GiB1.00 GiB23.91 GiB0.09 GiB6±25%
EVA-Qwen2.5-32B-v0.1Q5_K_L32.8B22.11 GiB1.00 GiB23.91 GiB0.09 GiB6±25%
cogito-v1-preview-qwen-32BQ5_K_L32.8B22.11 GiB1.00 GiB23.91 GiB0.09 GiB6±25%
QwQ-32B-Snowdrop-v0Q5_K_L32.8B22.11 GiB1.00 GiB23.91 GiB0.09 GiB6±25%
HarmonicHarlequin_v5-20BI1-Q3_K_M33.3B15.04 GiB8.13 GiB23.90 GiB0.10 GiB6±25%
Nemotron-Labs-Audex-30B-A3BQ4_K_M32.0B23.16 GiB0.00 GiB23.90 GiB0.10 GiB6±25%
spoomplesmaxx-v2.1-30BI1-Q6_K28.9B22.06 GiB1.00 GiB23.87 GiB0.13 GiB6±25%
Huihui-granite-4.1-30b-abliteratedI1-Q6_K28.9B22.06 GiB1.00 GiB23.87 GiB0.13 GiB6±25%
granite-4.1-30b-hereticI1-Q6_K28.9B22.06 GiB1.00 GiB23.87 GiB0.13 GiB6±25%
granite-4.1-30bQ6_K28.9B22.06 GiB1.00 GiB23.87 GiB0.13 GiB6±25%
OpenBuddy-R1-0528-Distill-Qwen3-32B-Preview0-QATQ5_K_L32.8B22.07 GiB1.00 GiB23.86 GiB0.14 GiB6±25%
KAT-DevQ5_K_L32.8B22.07 GiB1.00 GiB23.86 GiB0.14 GiB6±25%
Qwen3-VL-32B-InstructQ5_K_L33.4B22.07 GiB1.00 GiB23.86 GiB0.14 GiB6±25%
DeepSWE-PreviewQ5_K_L32.8B22.07 GiB1.00 GiB23.86 GiB0.14 GiB6±25%
Bielik-11B-v2.3-InstructQ8_011.2B22.34 GiB0.78 GiB23.86 GiB0.14 GiB6±25%
Huihui-Qwen3.5-35B-A3B-abliteratedMoEI1-Q5_K_M36.0B23.06 GiB0.08 GiB23.85 GiB0.15 GiB35±37%
Qwen3.5-35B-A3B-BaseMoEI1-Q5_K_M36.0B23.06 GiB0.08 GiB23.85 GiB0.15 GiB35±37%
Qwen3.5-35B-A3B-ultra-uncensored-hereticMoEQ5_K_M35.1B23.06 GiB0.08 GiB23.85 GiB0.15 GiB35±37%
Qwen3.5-35B-A3B-Claude-4.6-Opus-Reasoning-DistilledMoEI1-Q5_K_M36.0B23.06 GiB0.08 GiB23.85 GiB0.15 GiB35±37%
Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16UD-Q5_K_S33.0B23.10 GiB0.00 GiB23.85 GiB0.15 GiB6±25%
Qwen3.6-35B-A3B-uncensored-hereticMoEQ5_K_M35.1B23.06 GiB0.08 GiB23.85 GiB0.15 GiB35±37%
●From the file◐Predictedwhat these mean

Speed is modeled, not measured: decode is memory-bandwidth bound, so tokens per second is bytes read per token against achievable bandwidth. Mixture-of-experts models carry a wider band because only the routed experts are read each step, and few have been measured publicly.

Questions people ask

What AI models can a AMD Ryzen AI Max 385 (Radeon 8050S) run?
1985 of 2118 indexed open-weight models fit a AMD Ryzen AI Max 385 (Radeon 8050S) at 4,096 context with f16 KV cache, the largest being Qwen3-Coder-Next-REAM at I1-IQ3_XS. That covers text, vision-language, image, video and speech models.
How much usable memory does a AMD Ryzen AI Max 385 (Radeon 8050S) actually have?
Its nameplate is 32 GB, but about 22.32 GiB is available to a model once driver and compositor overhead is accounted for, and only 24 GB of the pool can be allocated to the GPU at all.
Is a AMD Ryzen AI Max 385 (Radeon 8050S) fast for local AI?
Its memory bandwidth is 256 GB/s, and that figure — not teraflops — is what governs token generation speed. Capacity decides what you can run; bandwidth decides how fast it runs.