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. 1967 of 2118 indexed models fit at 32K context with q4_0 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 1687vision language 176audio tts 21image 2video 16audio asr 39embedding 26

What fits at 32K context

largest quantization that fits, per model · 1967 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
llm-jp-4-32b-a3b-thinkingMoEQ5_K_M32.1B22.73 GiB0.56 GiB24.00 GiB0.00 GiB24±37%
magnum-v2-32bQ5_K_S32.5B20.92 GiB2.25 GiB23.97 GiB0.03 GiB6±25%
Nemotron-Cascade-2-30B-A3B-heretic-ara-uncensoredMoEI1-Q4_K_M31.6B22.83 GiB0.46 GiB23.97 GiB0.03 GiB27±37%
Nemotron-Cascade-2-30B-A3BMoEI1-Q4_K_M31.6B22.83 GiB0.46 GiB23.97 GiB0.03 GiB27±37%
GLM-4.7-Flash-REAP-23B-A3BMoEQ8_023.0B22.78 GiB0.46 GiB23.95 GiB0.05 GiB22±37%
GLM-4.7-Flash-REAP-23B-A3B-absolute-heresyMoEQ8_023.0B22.78 GiB0.46 GiB23.95 GiB0.05 GiB22±37%
Pantheon-Reasoning-27BQ6_K_L27.8B22.62 GiB0.56 GiB23.95 GiB0.05 GiB6±25%
Qwen3.5-27BQ6_K_L27.8B22.62 GiB0.56 GiB23.95 GiB0.05 GiB6±25%
Huihui-Qwen3.5-35B-A3B-abliteratedMoEI1-Q5_K_M36.0B23.06 GiB0.18 GiB23.94 GiB0.06 GiB34±37%
Qwen3.5-35B-A3B-BaseMoEI1-Q5_K_M36.0B23.06 GiB0.18 GiB23.94 GiB0.06 GiB34±37%
Qwen3.5-35B-A3B-ultra-uncensored-hereticMoEQ5_K_M35.1B23.06 GiB0.18 GiB23.94 GiB0.06 GiB34±37%
Qwen3.5-35B-A3B-Claude-4.6-Opus-Reasoning-DistilledMoEI1-Q5_K_M36.0B23.06 GiB0.18 GiB23.94 GiB0.06 GiB34±37%
Qwen3.6-35B-A3B-uncensored-hereticMoEQ5_K_M35.1B23.06 GiB0.18 GiB23.94 GiB0.06 GiB34±37%
Ornith-1.0-35B-uncensored-hereticMoEQ5_K_M35.1B23.06 GiB0.18 GiB23.94 GiB0.06 GiB34±37%
Nex-N2-mini-ultra-uncensored-hereticMoEQ5_K_M35.1B23.06 GiB0.18 GiB23.94 GiB0.06 GiB34±37%
command-r-35b-writer-v2I1-Q2_K_S35.0B11.86 GiB11.25 GiB23.92 GiB0.08 GiB6±25%
Carnice-Qwen3.6-MoE-35B-A3BMoEI1-Q5_K_M36.0B23.03 GiB0.18 GiB23.91 GiB0.09 GiB34±37%
Qwen35B-Agent-R2-AbliteratedMoEI1-Q5_K_M34.7B23.03 GiB0.18 GiB23.91 GiB0.09 GiB34±37%
Qwen3.6-35B-A3B-Claude-4.6-Opus-Reasoning-DistilledMoEI1-Q5_K_M36.0B23.03 GiB0.18 GiB23.91 GiB0.09 GiB34±37%
Darwin-35B-A3B-OpusMoEI1-Q5_K_M36.0B23.03 GiB0.18 GiB23.91 GiB0.09 GiB34±37%
Qwen35B-Agent-R2MoEI1-Q5_K_M34.7B23.03 GiB0.18 GiB23.91 GiB0.09 GiB34±37%
Carnice-MoE-35B-A3BMoEI1-Q5_K_M36.0B23.03 GiB0.18 GiB23.91 GiB0.09 GiB34±37%
spoomplesmaxx-flash-35B-A3MoEI1-Q5_K_M35.1B23.03 GiB0.18 GiB23.91 GiB0.09 GiB34±37%
Huihui-Qwen3.6-35B-A3B-Claude-4.7-Opus-abliteratedMoEI1-Q5_K_M36.0B23.03 GiB0.18 GiB23.91 GiB0.09 GiB34±37%
Qwen3.6-35B-A3B-Uncensored-AggressiveMoEI1-Q5_K_M35.1B23.03 GiB0.18 GiB23.91 GiB0.09 GiB34±37%
Holo-3.1-35B-A3BMoEQ5_K_M35.1B23.03 GiB0.18 GiB23.91 GiB0.09 GiB34±37%
WorldSim-Opus-3.6-35B-A3BMoEI1-Q5_K_M35.1B23.03 GiB0.18 GiB23.91 GiB0.09 GiB34±37%
Qwen3.6-35B-A3B-abliterated-MAXMoEI1-Q5_K_M35.1B23.03 GiB0.18 GiB23.91 GiB0.09 GiB34±37%
Huihui-Qwen3.6-35B-A3B-abliteratedMoEI1-Q5_K_M36.0B23.03 GiB0.18 GiB23.91 GiB0.09 GiB34±37%
Qwopus3.6-35B-A3B-v1MoEI1-Q5_K_M36.0B23.03 GiB0.18 GiB23.91 GiB0.09 GiB34±37%
Qwen3.6-35B-A3B-StyleTuneMoEI1-Q5_K_M35.1B23.03 GiB0.18 GiB23.91 GiB0.09 GiB34±37%
Qwen3.6-35B-A3B-abliteratedMoEI1-Q5_K_M35.1B23.03 GiB0.18 GiB23.91 GiB0.09 GiB34±37%
Qwen3.6-35B-A3B-Kimi-K2.6-Reasoning-DistilledMoEQ5_K_M36.0B23.03 GiB0.18 GiB23.91 GiB0.09 GiB34±37%
Qwen3.6-35B-A3B-Claude-4.7-Opus-Reasoning-DistilledMoEQ5_K_M36.0B23.03 GiB0.18 GiB23.91 GiB0.09 GiB34±37%
Qwen3.6-35B-A3B-hereticMoEQ5_K_M35.1B23.03 GiB0.18 GiB23.91 GiB0.09 GiB34±37%
Qwen3.6-35B-A3B-abliterated-v4MoEQ5_K_M34.7B23.03 GiB0.18 GiB23.91 GiB0.09 GiB34±37%
Qwen3.6-35B-A3B-java-v1MoEQ5_K_M34.7B23.03 GiB0.18 GiB23.91 GiB0.09 GiB34±37%
Huihui-Qwen3.5-35B-A3B-Claude-4.6-Opus-abliteratedMoEQ5_K_M36.0B23.03 GiB0.18 GiB23.91 GiB0.09 GiB34±37%
0GM-1.0-35B-A3B-0427MoEI1-Q5_K_M36.0B23.03 GiB0.18 GiB23.91 GiB0.09 GiB34±37%
HopCoder-Mini-35B-A3B-VL36MoEQ5_K_M35.1B23.03 GiB0.18 GiB23.91 GiB0.09 GiB34±37%
Phi-3.5-MoE-instructMoEKV unresolvedQ4_041.9B22.08 GiB1.13 GiB23.91 GiB0.09 GiB16±37%
Qwen3-53B-A3B-2507-THINKING-TOTAL-RECALL-v2-MASTER-CODERMoEI1-IQ3_M53.0B21.74 GiB1.48 GiB23.91 GiB0.09 GiB18±37%
Nemotron-Labs-Audex-30B-A3BQ4_K_M32.0B23.16 GiB0.00 GiB23.90 GiB0.10 GiB6±25%
Qwen3.5-40B-RoughHouse-Claude-4.6-Opus-Polar-Deckard-Uncensored-Heretic-ThinkingI1-Q4_K_M39.5B22.29 GiB0.84 GiB23.90 GiB0.10 GiB6±25%
Qwen3.5-40B-Claude-4.5-Opus-High-Reasoning-Thinking-uncensored-hereticQ4_K_M39.5B22.29 GiB0.84 GiB23.90 GiB0.10 GiB6±25%
Gemma4-Gutenberg-31BQ5_K_L31.3B21.37 GiB1.74 GiB23.89 GiB0.11 GiB6±25%
gemma-4-31B-itQ5_K_L31.3B21.37 GiB1.74 GiB23.89 GiB0.11 GiB6±25%
Gemma4-Gutenberg-31B-HereticQ5_K_L31.3B21.37 GiB1.74 GiB23.89 GiB0.11 GiB6±25%
Equinox-31BQ5_K_L31.3B21.37 GiB1.74 GiB23.89 GiB0.11 GiB6±25%
gemma-4-31B-it-SDFT-Heretic-RPQ5_K_L30.7B21.37 GiB1.74 GiB23.89 GiB0.11 GiB6±25%
Qwen3.5-40B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-ThinkingI1-Q4_K_M39.5B22.28 GiB0.84 GiB23.89 GiB0.11 GiB6±25%
codegeex4-all-9bF169.4B17.52 GiB5.63 GiB23.89 GiB0.11 GiB6±25%
L3-DARKEST-PLANET-16.5BQ2_K16.5B20.65 GiB2.50 GiB23.89 GiB0.11 GiB6±25%
glm-4-9b-chat-abliteratedF169.4B17.52 GiB5.63 GiB23.89 GiB0.11 GiB6±25%
glm-4-9b-chatBF169.4B17.52 GiB5.63 GiB23.89 GiB0.11 GiB6±25%
v6-Finch-14B-HFQ8_014.1B14.55 GiB8.58 GiB23.87 GiB0.13 GiB6±25%
Yi-1.5-6B-ChatF326.1B22.58 GiB0.56 GiB23.87 GiB0.13 GiB6±25%
Seed-OSS-36B-InstructQ4_K_L36.2B20.82 GiB2.25 GiB23.87 GiB0.13 GiB6±25%
Hermes-4.3-36BQ4_K_L36.2B20.82 GiB2.25 GiB23.87 GiB0.13 GiB6±25%
Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16UD-Q5_K_S33.0B23.10 GiB0.00 GiB23.85 GiB0.15 GiB6±25%
●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?
1967 of 2118 indexed open-weight models fit a AMD Ryzen AI Max 385 (Radeon 8050S) at 32,768 context with q4_0 KV cache, the largest being llm-jp-4-32b-a3b-thinking at Q5_K_M. 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.