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. 1954 of 2118 indexed models fit at 64K 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
vision language 174text 1676audio tts 21audio asr 39image 2video 16embedding 26

What fits at 64K context

largest quantization that fits, per model · 1954 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Qwen3.6-40B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-ThinkingIQ4_NL39.5B21.55 GiB1.69 GiB24.00 GiB0.00 GiB6±25%
Salience-1.5-ProMoEQ5_K_S36.0B22.94 GiB0.35 GiB24.00 GiB0.00 GiB32±37%
Qwable-v1MoEQ5_K_S36.0B22.94 GiB0.35 GiB24.00 GiB0.00 GiB32±37%
T-SearchMoEQ5_K_S36.0B22.94 GiB0.35 GiB24.00 GiB0.00 GiB32±37%
dolphin-2.6-mixtral-8x7bMoEI1-Q3_K_M46.7B21.00 GiB2.25 GiB23.98 GiB0.02 GiB9±37%
Nous-Hermes-2-Mixtral-8x7B-DPOMoEQ3_K_M46.7B21.00 GiB2.25 GiB23.98 GiB0.02 GiB9±37%
Mixtral-8x7B-Instruct-v0.1MoEQ3_K_M46.7B21.00 GiB2.25 GiB23.98 GiB0.02 GiB9±37%
xLAM-8x7b-rMoEQ3_K_M46.7B21.00 GiB2.25 GiB23.98 GiB0.02 GiB9±37%
dolphin-2.5-mixtral-8x7bMoEQ3_K_M46.7B21.00 GiB2.25 GiB23.98 GiB0.02 GiB9±37%
Mixtral-8x7B-v0.1MoEQ3_K_M46.7B21.00 GiB2.25 GiB23.98 GiB0.02 GiB9±37%
14BQ5_014.2B9.17 GiB14.06 GiB23.98 GiB0.02 GiB6±25%
Qwen3.6-35B-A3B-REAM-160-ru-agentMoEQ8_023.6B22.92 GiB0.35 GiB23.98 GiB0.02 GiB27±37%
NSFW_13B_sftQ5_K_M13.3B9.17 GiB14.06 GiB23.98 GiB0.02 GiB6±25%
Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-PreservedMoEQ5_K_S35.1B22.91 GiB0.35 GiB23.97 GiB0.03 GiB32±37%
Qwen3.5-35B-A3B-uncensored-heretic-v2-Native-MTP-PreservedMoEQ5_K_S35.1B22.91 GiB0.35 GiB23.96 GiB0.04 GiB32±37%
Nemotron-Cascade-2-30B-A3BMoEQ5_K_S31.6B22.36 GiB0.91 GiB23.96 GiB0.04 GiB24±37%
Gemma4-Gutenberg-31BQ5_K_S31.3B20.03 GiB3.14 GiB23.95 GiB0.05 GiB6±25%
gemma-4-31B-itQ5_K_S31.3B20.03 GiB3.14 GiB23.95 GiB0.05 GiB6±25%
Gemma4-Gutenberg-31B-HereticQ5_K_S31.3B20.03 GiB3.14 GiB23.95 GiB0.05 GiB6±25%
Equinox-31BQ5_K_S31.3B20.03 GiB3.14 GiB23.95 GiB0.05 GiB6±25%
gemma-4-31B-it-SDFT-Heretic-RPQ5_K_S30.7B20.03 GiB3.14 GiB23.95 GiB0.05 GiB6±25%
Qwen3.6-35B-A3B-NVFP4-FastMoENVFP421.8B22.89 GiB0.35 GiB23.95 GiB0.05 GiB32±37%
Pantheon-Reasoning-27BQ6_K27.8B22.05 GiB1.13 GiB23.94 GiB0.06 GiB6±25%
Qwen3.5-27BQ6_K27.8B22.05 GiB1.13 GiB23.94 GiB0.06 GiB6±25%
Llama3.2-30B-A3B-II-Dark-Champion-INSTRUCT-Heretic-Abliterated-UncensoredMoEI1-Q5_K_M30.0B19.92 GiB3.30 GiB23.94 GiB0.06 GiB12±37%
Qwen3.6-35B-A3B-Fable-5-DistillMoEI1-Q5_K_S36.0B22.88 GiB0.35 GiB23.93 GiB0.07 GiB32±37%
Qwable-v2MoEI1-Q5_K_S36.0B22.88 GiB0.35 GiB23.93 GiB0.07 GiB32±37%
Qwen3.6-35B-A3B-YOYO-V2MoEI1-Q5_K_S36.0B22.88 GiB0.35 GiB23.93 GiB0.07 GiB32±37%
Ornith-1.0-35B-FP8-BLOCK-MTPMoEI1-Q5_K_S35.5B22.88 GiB0.35 GiB23.93 GiB0.07 GiB32±37%
fable-coder-35B-A3BMoEI1-Q5_K_S36.0B22.88 GiB0.35 GiB23.93 GiB0.07 GiB32±37%
Qwen3.6-35B-A3B-AntiLoopMoEI1-Q5_K_S36.0B22.88 GiB0.35 GiB23.93 GiB0.07 GiB32±37%
PINQWEN-3.6-35B-CLEAN-BF16MoEI1-Q5_K_S36.0B22.88 GiB0.35 GiB23.93 GiB0.07 GiB32±37%
UniMath-35B-A3BMoEI1-Q5_K_S36.0B22.88 GiB0.35 GiB23.93 GiB0.07 GiB32±37%
Ornith-1.0-35B-Heretic-MTPMoEI1-Q5_K_S—22.88 GiB0.35 GiB23.93 GiB0.07 GiB32±37%
Fawen-1.0-35BMoEI1-Q5_K_S36.0B22.88 GiB0.35 GiB23.93 GiB0.07 GiB32±37%
CyberStrike-OffSec-35BMoEQ5_K_S35.1B22.88 GiB0.35 GiB23.93 GiB0.07 GiB32±37%
Qwen3.6-35B-A3BMoEQ5_K_S36.0B22.88 GiB0.35 GiB23.93 GiB0.07 GiB32±37%
gemma-3-12b-it-abliterated-v2F1611.8B21.92 GiB1.26 GiB23.93 GiB0.07 GiB6±25%
gemma-3-12b-it-abliteratedBF1612.2B21.92 GiB1.26 GiB23.93 GiB0.07 GiB6±25%
gemma-3-12b-itBF1612.2B21.92 GiB1.26 GiB23.93 GiB0.07 GiB6±25%
Open_Gpt4_8x7B_v0.2MoEQ3_K_M46.7B20.93 GiB2.25 GiB23.92 GiB0.08 GiB9±37%
Laguna-XS-2.1MoEQ5_K_L33.4B22.47 GiB0.74 GiB23.91 GiB0.09 GiB28±37%
GLM-Z1-32B-0414Q5_K_L32.6B22.05 GiB1.07 GiB23.91 GiB0.09 GiB6±25%
GLM-4-32B-0414Q5_K_L32.6B22.05 GiB1.07 GiB23.91 GiB0.09 GiB6±25%
NVIDIA-Nemotron-3-Nano-30B-A3B-BF16MoEQ5_K_S31.6B22.31 GiB0.91 GiB23.90 GiB0.10 GiB24±37%
Qwen3-53B-A3B-2507-THINKING-TOTAL-RECALL-v2-MASTER-CODERMoEI1-IQ3_XS53.0B20.26 GiB2.95 GiB23.90 GiB0.10 GiB14±37%
Nemotron-Labs-Audex-30B-A3BQ4_K_M32.0B23.16 GiB0.00 GiB23.90 GiB0.10 GiB6±25%
OLMo-2-1124-13B-InstructQ5_K_M13.7B9.09 GiB14.06 GiB23.90 GiB0.10 GiB6±25%
Hunyuan-A13B-InstructMoEUD-TQ1_080.4B20.95 GiB2.25 GiB23.90 GiB0.10 GiB6±25%
Gemma-4-Novelist-Eclipse-31BQ4_K_L32.7B19.96 GiB3.14 GiB23.89 GiB0.11 GiB6±25%
Gemma-4-31B-StyleTuneQ4_K_L32.7B19.96 GiB3.14 GiB23.89 GiB0.11 GiB6±25%
GPT-NeoX-20B-ErebusI1-IQ1_M20.6B4.51 GiB18.56 GiB23.86 GiB0.14 GiB6±25%
spoomplesmaxx-v2.1-30BI1-Q5_K_S28.9B18.54 GiB4.50 GiB23.85 GiB0.15 GiB6±25%
Huihui-granite-4.1-30b-abliteratedI1-Q5_K_S28.9B18.54 GiB4.50 GiB23.85 GiB0.15 GiB6±25%
granite-4.1-30b-hereticI1-Q5_K_S28.9B18.54 GiB4.50 GiB23.85 GiB0.15 GiB6±25%
granite-4.1-30bQ5_028.9B18.54 GiB4.50 GiB23.85 GiB0.15 GiB6±25%
Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16UD-Q5_K_S33.0B23.10 GiB0.00 GiB23.85 GiB0.15 GiB6±25%
granite-34b-code-base-8kI1-Q5_K_M33.7B23.05 GiB0.00 GiB23.83 GiB0.17 GiB6±25%
internlm2-math-plus-20bQ8_019.9B19.66 GiB3.38 GiB23.79 GiB0.21 GiB6±25%
CallerQ4_K_M32.8B18.49 GiB4.50 GiB23.79 GiB0.21 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?
1954 of 2118 indexed open-weight models fit a AMD Ryzen AI Max 385 (Radeon 8050S) at 65,536 context with q4_0 KV cache, the largest being Qwen3.6-40B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-Thinking at IQ4_NL. 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.