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 16K context with q8_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 16K context

largest quantization that fits, per model · 1967 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
NVIDIA-Nemotron-3-Nano-30B-A3B-BF16MoEQ4_K_M31.6B22.89 GiB0.43 GiB24.00 GiB0.00 GiB27±37%
OpenBuddy-R1-0528-Distill-Qwen3-32B-Preview0-QATQ5_K_S32.8B21.08 GiB2.13 GiB24.00 GiB0.00 GiB6±25%
Qwen3-VL-32B-Instruct-ultra-uncensored-hereticI1-Q5_K_S33.4B21.08 GiB2.13 GiB24.00 GiB0.00 GiB6±25%
Huihui-Qwen3-VL-32B-Instruct-abliteratedI1-Q5_K_S33.4B21.08 GiB2.13 GiB24.00 GiB0.00 GiB6±25%
KAT-DevQ5_K_S32.8B21.08 GiB2.13 GiB24.00 GiB0.00 GiB6±25%
ColorGUI-32BI1-Q5_K_S33.4B21.08 GiB2.13 GiB24.00 GiB0.00 GiB6±25%
Qwen3-VL-32B-InstructQ5_K_S33.4B21.08 GiB2.13 GiB24.00 GiB0.00 GiB6±25%
Qwen3-VL-32B-ThinkingQ5_K_S33.4B21.08 GiB2.13 GiB24.00 GiB0.00 GiB6±25%
Qwen3-32B-UncensoredI1-Q5_K_S32.8B21.08 GiB2.13 GiB24.00 GiB0.00 GiB6±25%
Qwen3-32BQ5_K_S32.8B21.08 GiB2.13 GiB24.00 GiB0.00 GiB6±25%
Qwen3-32B-abliteratedI1-Q5_K_S32.8B21.08 GiB2.13 GiB24.00 GiB0.00 GiB6±25%
DeepSWE-PreviewQ5_K_S32.8B21.08 GiB2.13 GiB24.00 GiB0.00 GiB6±25%
AReaL-boba-2-32BI1-Q5_K_S32.8B21.08 GiB2.13 GiB24.00 GiB0.00 GiB6±25%
Assistant_Pepe_32BI1-Q5_K_S32.8B21.08 GiB2.13 GiB24.00 GiB0.00 GiB6±25%
Gemma-4-31B-Isometry-RPI1-Q5_K_M32.7B21.25 GiB1.95 GiB23.98 GiB0.02 GiB6±25%
Gemma-4-Dark-Gemistry-31BI1-Q5_K_M32.7B21.25 GiB1.95 GiB23.98 GiB0.02 GiB6±25%
Prosopon-31BI1-Q5_K_M32.7B21.25 GiB1.95 GiB23.98 GiB0.02 GiB6±25%
Gemma-4-Novelist-Eclipse-31BI1-Q5_K_M32.7B21.25 GiB1.95 GiB23.98 GiB0.02 GiB6±25%
Giftige-Blume-31B-v1-StyleSwapI1-Q5_K_M32.7B21.25 GiB1.95 GiB23.98 GiB0.02 GiB6±25%
G4-MeroMero-31B-StyleSwapI1-Q5_K_M32.7B21.25 GiB1.95 GiB23.98 GiB0.02 GiB6±25%
Gemma-4-31B-StyleTune-heretic-araI1-Q5_K_M32.7B21.25 GiB1.95 GiB23.98 GiB0.02 GiB6±25%
Pantheon-Reasoning-31B-1.1I1-Q5_K_M32.7B21.25 GiB1.95 GiB23.98 GiB0.02 GiB6±25%
Gemma-4-31B-StyleTuneI1-Q5_K_M32.7B21.25 GiB1.95 GiB23.98 GiB0.02 GiB6±25%
Barcenas-StyleTune-31B-FableI1-Q5_K_M32.1B21.25 GiB1.95 GiB23.98 GiB0.02 GiB6±25%
llm-jp-4-32b-a3b-thinkingMoEQ5_K_M32.1B22.73 GiB0.53 GiB23.97 GiB0.03 GiB24±37%
Skyfall-31B-v4.2Q5_K_L31.4B21.35 GiB1.79 GiB23.96 GiB0.04 GiB6±25%
Hunyuan-A13B-InstructMoEIQ2_S80.4B22.19 GiB1.06 GiB23.95 GiB0.05 GiB6±25%
Nemotron-Cascade-2-30B-A3B-heretic-ara-uncensoredMoEI1-Q4_K_M31.6B22.83 GiB0.43 GiB23.94 GiB0.06 GiB27±37%
Nemotron-Cascade-2-30B-A3BMoEI1-Q4_K_M31.6B22.83 GiB0.43 GiB23.94 GiB0.06 GiB27±37%
Phi-3.5-MoE-instructMoEKV unresolvedQ4_K_S41.9B22.18 GiB1.06 GiB23.94 GiB0.06 GiB16±37%
Huihui-Qwen3.5-35B-A3B-abliteratedMoEI1-Q5_K_M36.0B23.06 GiB0.17 GiB23.93 GiB0.07 GiB34±37%
Qwen3.5-35B-A3B-BaseMoEI1-Q5_K_M36.0B23.06 GiB0.17 GiB23.93 GiB0.07 GiB34±37%
Qwen3.5-35B-A3B-ultra-uncensored-hereticMoEQ5_K_M35.1B23.06 GiB0.17 GiB23.93 GiB0.07 GiB34±37%
Qwen3.5-35B-A3B-Claude-4.6-Opus-Reasoning-DistilledMoEI1-Q5_K_M36.0B23.06 GiB0.17 GiB23.93 GiB0.07 GiB34±37%
Qwen3.6-35B-A3B-uncensored-hereticMoEQ5_K_M35.1B23.06 GiB0.17 GiB23.93 GiB0.07 GiB34±37%
Ornith-1.0-35B-uncensored-hereticMoEQ5_K_M35.1B23.06 GiB0.17 GiB23.93 GiB0.07 GiB34±37%
Nex-N2-mini-ultra-uncensored-hereticMoEQ5_K_M35.1B23.06 GiB0.17 GiB23.93 GiB0.07 GiB34±37%
GLM-4.7-Flash-REAP-23B-A3BMoEQ8_023.0B22.78 GiB0.44 GiB23.93 GiB0.07 GiB22±37%
GLM-4.7-Flash-REAP-23B-A3B-absolute-heresyMoEQ8_023.0B22.78 GiB0.44 GiB23.93 GiB0.07 GiB22±37%
Pantheon-Reasoning-27BQ6_K_L27.8B22.62 GiB0.53 GiB23.92 GiB0.08 GiB6±25%
Qwen3.5-27BQ6_K_L27.8B22.62 GiB0.53 GiB23.92 GiB0.08 GiB6±25%
Nemotron-Labs-Audex-30B-A3BQ4_K_M32.0B23.16 GiB0.00 GiB23.90 GiB0.10 GiB6±25%
Carnice-Qwen3.6-MoE-35B-A3BMoEI1-Q5_K_M36.0B23.03 GiB0.17 GiB23.90 GiB0.10 GiB34±37%
Qwen35B-Agent-R2-AbliteratedMoEI1-Q5_K_M34.7B23.03 GiB0.17 GiB23.90 GiB0.10 GiB34±37%
Qwen3.6-35B-A3B-Claude-4.6-Opus-Reasoning-DistilledMoEI1-Q5_K_M36.0B23.03 GiB0.17 GiB23.90 GiB0.10 GiB34±37%
Darwin-35B-A3B-OpusMoEI1-Q5_K_M36.0B23.03 GiB0.17 GiB23.90 GiB0.10 GiB34±37%
Qwen35B-Agent-R2MoEI1-Q5_K_M34.7B23.03 GiB0.17 GiB23.90 GiB0.10 GiB34±37%
Carnice-MoE-35B-A3BMoEI1-Q5_K_M36.0B23.03 GiB0.17 GiB23.90 GiB0.10 GiB34±37%
spoomplesmaxx-flash-35B-A3MoEI1-Q5_K_M35.1B23.03 GiB0.17 GiB23.90 GiB0.10 GiB34±37%
Huihui-Qwen3.6-35B-A3B-Claude-4.7-Opus-abliteratedMoEI1-Q5_K_M36.0B23.03 GiB0.17 GiB23.90 GiB0.10 GiB34±37%
Qwen3.6-35B-A3B-Uncensored-AggressiveMoEI1-Q5_K_M35.1B23.03 GiB0.17 GiB23.90 GiB0.10 GiB34±37%
Holo-3.1-35B-A3BMoEQ5_K_M35.1B23.03 GiB0.17 GiB23.90 GiB0.10 GiB34±37%
WorldSim-Opus-3.6-35B-A3BMoEI1-Q5_K_M35.1B23.03 GiB0.17 GiB23.90 GiB0.10 GiB34±37%
Qwen3.6-35B-A3B-abliterated-MAXMoEI1-Q5_K_M35.1B23.03 GiB0.17 GiB23.90 GiB0.10 GiB34±37%
Huihui-Qwen3.6-35B-A3B-abliteratedMoEI1-Q5_K_M36.0B23.03 GiB0.17 GiB23.90 GiB0.10 GiB34±37%
Qwopus3.6-35B-A3B-v1MoEI1-Q5_K_M36.0B23.03 GiB0.17 GiB23.90 GiB0.10 GiB34±37%
Qwen3.6-35B-A3B-StyleTuneMoEI1-Q5_K_M35.1B23.03 GiB0.17 GiB23.90 GiB0.10 GiB34±37%
Qwen3.6-35B-A3B-abliteratedMoEI1-Q5_K_M35.1B23.03 GiB0.17 GiB23.90 GiB0.10 GiB34±37%
Qwen3.6-35B-A3B-Kimi-K2.6-Reasoning-DistilledMoEQ5_K_M36.0B23.03 GiB0.17 GiB23.90 GiB0.10 GiB34±37%
Qwen3.6-35B-A3B-Claude-4.7-Opus-Reasoning-DistilledMoEQ5_K_M36.0B23.03 GiB0.17 GiB23.90 GiB0.10 GiB34±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?
1967 of 2118 indexed open-weight models fit a AMD Ryzen AI Max 385 (Radeon 8050S) at 16,384 context with q8_0 KV cache, the largest being NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 at Q4_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.