AMD · unified x86

AMD Ryzen AI Max 385 (Radeon 8050S)

AMD Ryzen AI Max 385 (Radeon 8050S) has 64 GB of VRAM at 256 GB/s — about 44.64 GiB usable after driver and compositor overhead. 2049 of 2118 indexed models fit at 16K context with q8_0 KV. Note only 48 GB of its 64 GB is allocatable to the GPU.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
64 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 1760image 2vision language 185audio asr 39audio tts 21video 16embedding 26

What fits at 16K context

largest quantization that fits, per model · 2049 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Llama-3_1-Nemotron-51B-InstructIQ4_XS51.5B25.83 GiB21.25 GiB47.92 GiB0.08 GiB3±25%
Step-3.5-Flash-REAP-121B-A11BI1-IQ3_XXS121B43.40 GiB3.74 GiB47.86 GiB0.14 GiB3±25%
Qwen3-Coder-REAP-25B-A3BMoEBF1624.9B46.34 GiB0.80 GiB47.83 GiB0.17 GiB11±37%
Hunyuan-A13B-InstructMoEQ4_K_L80.4B46.05 GiB1.06 GiB47.81 GiB0.19 GiB3±25%
HunyuanImage-2.1Q5_017.5B47.04 GiB0.00 GiB47.79 GiB0.21 GiB3±25%
Tess-3-Mistral-Nemo-12BF3212.2B45.63 GiB1.33 GiB47.71 GiB0.29 GiB3±25%
Lumimaid-v0.2-12BF3212.2B45.63 GiB1.33 GiB47.71 GiB0.29 GiB3±25%
MN-Violet-Lotus-12BF3212.2B45.63 GiB1.33 GiB47.71 GiB0.29 GiB3±25%
Mistral-Nemo-Instruct-2407F3212.2B45.63 GiB1.33 GiB47.71 GiB0.29 GiB3±25%
MN-12B-Celeste-V1.9F3212.2B45.63 GiB1.33 GiB47.71 GiB0.29 GiB3±25%
magnum-v2.5-12b-ktoF3212.2B45.63 GiB1.33 GiB47.71 GiB0.29 GiB3±25%
magnum-v2-12bF3212.2B45.63 GiB1.33 GiB47.71 GiB0.29 GiB3±25%
CodeLlama-70b-Instruct-hfI1-Q5_K_S69.0B44.20 GiB2.66 GiB47.68 GiB0.32 GiB3±25%
CodeLlama-70b-Python-hfI1-Q5_K_S69.0B44.20 GiB2.66 GiB47.68 GiB0.32 GiB3±25%
Nous-Hermes-Llama2-70bI1-Q5_K_S69.0B44.20 GiB2.66 GiB47.68 GiB0.32 GiB3±25%
Midnight-Miqu-70B-v1.5I1-Q5_K_S69.0B44.20 GiB2.66 GiB47.68 GiB0.32 GiB3±25%
KafkaLM-70B-German-V0.1Q5_069.0B44.20 GiB2.66 GiB47.68 GiB0.32 GiB3±25%
llama2_70b_chat_uncensoredQ5_069.0B44.20 GiB2.66 GiB47.68 GiB0.32 GiB3±25%
Xwin-LM-70b-V0.1Q5_069.0B44.20 GiB2.66 GiB47.68 GiB0.32 GiB3±25%
Llama-2-70b-chat-hfQ5_069.0B44.20 GiB2.66 GiB47.68 GiB0.32 GiB3±25%
Qwen3.5-122B-A10B-hereticMoEI1-IQ3_XS123B46.72 GiB0.20 GiB47.65 GiB0.35 GiB18±37%
Rombo-LLM-V3.0-Qwen-72bI1-Q4_K_M72.7B44.16 GiB2.66 GiB47.65 GiB0.35 GiB3±25%
Qwen2.5-72B-Instruct-abliteratedI1-Q4_K_M72.7B44.16 GiB2.66 GiB47.65 GiB0.35 GiB3±25%
Qwen2.5-72B-Instruct-abliterated-v2I1-Q4_K_M72.7B44.16 GiB2.66 GiB47.65 GiB0.35 GiB3±25%
HuatuoGPT-o1-72BQ4_K_M72.7B44.16 GiB2.66 GiB47.65 GiB0.35 GiB3±25%
MiroThinker-v1.0-72BI1-Q4_K_M72.7B44.16 GiB2.66 GiB47.65 GiB0.35 GiB3±25%
EVA-Qwen2.5-72B-v0.2Q4_K_M72.7B44.16 GiB2.66 GiB47.65 GiB0.35 GiB3±25%
Qwen2.5-Math-72B-InstructQ4_K_M72.7B44.16 GiB2.66 GiB47.65 GiB0.35 GiB3±25%
Qwen2.5-72B-InstructQ4_K_M72.7B44.16 GiB2.66 GiB47.65 GiB0.35 GiB3±25%
Malaysian-Qwen2.5-72B-InstructI1-Q4_K_M72.7B44.16 GiB2.66 GiB47.65 GiB0.35 GiB3±25%
Qwen2.5-72BI1-Q4_K_M72.7B44.16 GiB2.66 GiB47.65 GiB0.35 GiB3±25%
magnum-v4-72bI1-Q4_K_M72.7B44.16 GiB2.66 GiB47.65 GiB0.35 GiB3±25%
KAT-Dev-72B-ExpQ4_K_M72.7B44.16 GiB2.66 GiB47.65 GiB0.35 GiB3±25%
Homer-v1.0-Qwen2.5-72BQ4_K_M72.7B44.16 GiB2.66 GiB47.65 GiB0.35 GiB3±25%
Chuluun-Qwen2.5-72B-v0.01Q4_K_M72.7B44.16 GiB2.66 GiB47.65 GiB0.35 GiB3±25%
Qwen2.5-VL-72B-InstructQ4_K_M73.4B44.16 GiB2.66 GiB47.65 GiB0.35 GiB3±25%
Tower-Plus-72B-ultra-uncensored-hereticI1-Q4_K_M72.7B44.16 GiB2.66 GiB47.65 GiB0.35 GiB3±25%
Chronos-Platinum-72BQ4_K_M72.7B44.16 GiB2.66 GiB47.65 GiB0.35 GiB3±25%
UI-TARS-72B-DPOQ4_K_M73.4B44.16 GiB2.66 GiB47.65 GiB0.35 GiB3±25%
GLM-4.5-Air-REAP-82B-A12BMoEIQ4_NL81.9B45.37 GiB1.53 GiB47.62 GiB0.38 GiB10±37%
Qwen3-72B-SynthesisQ4_K_M72.7B44.12 GiB2.66 GiB47.60 GiB0.40 GiB3±25%
Behemoth-X-123B-v2IQ3_XXS123B43.78 GiB2.92 GiB47.56 GiB0.44 GiB3±25%
Mistral-Large-Instruct-2411IQ3_XXS123B43.78 GiB2.92 GiB47.56 GiB0.44 GiB3±25%
Qwen3.5-88BMoEI1-Q4_K_S87.7B46.63 GiB0.20 GiB47.55 GiB0.45 GiB16±37%
Huihui-Qwen3-Coder-Next-abliteratedMoEQ4_179.7B46.65 GiB0.20 GiB47.54 GiB0.46 GiB20±37%
Ornith-1.0-35B-AEON-Ultimate-Uncensored-BF16MoEQ5_K_M35.1B46.64 GiB0.17 GiB47.51 GiB0.49 GiB18±37%
c4ai-command-r-plus-08-2024IQ3_M104B44.41 GiB2.13 GiB47.41 GiB0.59 GiB3±25%
CalmeRys-78B-Orpo-v0.1I1-Q4_K_S78.0B43.72 GiB2.86 GiB47.41 GiB0.59 GiB3±25%
calme-2.3-rys-78bQ4_K_S78.0B43.72 GiB2.86 GiB47.41 GiB0.59 GiB3±25%
Qwen3-Coder-NextMoEUD-Q4_K_M79.7B45.92 GiB0.80 GiB47.40 GiB0.60 GiB18±37%
Mistral-Small-4-119B-2603MoEIQ3_XXS119B46.44 GiB0.19 GiB47.36 GiB0.64 GiB18±37%
Voxtral-Small-24B-2507F1624.3B45.21 GiB1.33 GiB47.36 GiB0.64 GiB3±25%
Laguna-S-2.1MoEIQ3_XXS118B46.07 GiB0.47 GiB47.27 GiB0.73 GiB16±37%
GLM-4.6VMoEIQ3_XXS108B44.99 GiB1.53 GiB47.25 GiB0.75 GiB12±37%
deepseek-llm-67b-chatI1-Q5_K_S67.4B43.28 GiB3.15 GiB47.24 GiB0.76 GiB3±25%
deepseek-llm-67b-baseI1-Q5_K_S67.4B43.28 GiB3.15 GiB47.24 GiB0.76 GiB3±25%
openbuddy-deepseek-67b-v15.3-4kI1-Q5_K_S67.4B43.28 GiB3.15 GiB47.24 GiB0.76 GiB3±25%
Devstral-2-123B-Instruct-2512Q2_K125B43.39 GiB2.92 GiB47.17 GiB0.83 GiB3±25%
Mistral-Medium-3.5-128BI1-Q2_K128B43.39 GiB2.92 GiB47.17 GiB0.83 GiB3±25%
XORTRON-NXTXPRTXXLI1-Q2_K128B43.39 GiB2.92 GiB47.17 GiB0.83 GiB3±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?
2049 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 Llama-3_1-Nemotron-51B-Instruct at IQ4_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 64 GB, but about 44.64 GiB is available to a model once driver and compositor overhead is accounted for, and only 48 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.