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. 2030 of 2118 indexed models fit at 64K 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 1742image 2vision language 184audio tts 21video 16audio asr 39embedding 26

What fits at 64K context

largest quantization that fits, per model · 2030 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Qwen3.5-88BMoEI1-Q4_087.7B46.41 GiB0.80 GiB47.94 GiB0.06 GiB14±37%
Mistral-Small-4-119B-2603MoEIQ3_XXS119B46.44 GiB0.75 GiB47.92 GiB0.08 GiB16±37%
GPT-NeoX-20B-ErebusI1-Q4_120.6B12.03 GiB35.06 GiB47.89 GiB0.11 GiB3±25%
Noromaid-20b-v0.1.1I1-IQ2_S20.0B5.96 GiB41.17 GiB47.87 GiB0.13 GiB3±25%
command-a-plus-05-2026-bf16MoEIQ1_S219B45.87 GiB1.29 GiB47.86 GiB0.14 GiB12±37%
Nous-Hermes-Llama2-70bQ4_K_S69.0B36.39 GiB10.63 GiB47.84 GiB0.16 GiB3±25%
llama2_70b_chat_uncensoredQ4_K_S69.0B36.39 GiB10.63 GiB47.84 GiB0.16 GiB3±25%
Xwin-LM-70b-V0.1Q4_K_S69.0B36.39 GiB10.63 GiB47.84 GiB0.16 GiB3±25%
Llama-2-70b-chat-hfQ4_K_S69.0B36.39 GiB10.63 GiB47.84 GiB0.16 GiB3±25%
Huihui-GLM-4.7-Flash-abliterated-57BMoEI1-Q6_K57.3B42.64 GiB4.45 GiB47.83 GiB0.17 GiB8±37%
CodeLlama-70b-Instruct-hfI1-Q4_069.0B36.34 GiB10.63 GiB47.79 GiB0.21 GiB3±25%
CodeLlama-70b-Python-hfI1-Q4_069.0B36.34 GiB10.63 GiB47.79 GiB0.21 GiB3±25%
Midnight-Miqu-70B-v1.5I1-Q4_069.0B36.34 GiB10.63 GiB47.79 GiB0.21 GiB3±25%
HunyuanImage-2.1Q5_017.5B47.04 GiB0.00 GiB47.79 GiB0.21 GiB3±25%
Mixtral_34Bx2_MoE_60BMoEQ5_K_M60.8B39.03 GiB7.97 GiB47.78 GiB0.22 GiB2±37%
Delphi-25B-SimpleRL-MathI1-Q3_K_M25.0B11.39 GiB35.56 GiB47.73 GiB0.27 GiB3±25%
Hunyuan-A13B-InstructMoEIQ4_NL80.4B42.77 GiB4.25 GiB47.71 GiB0.29 GiB3±25%
Qwen3-Coder-Next-REAMMoEI1-Q6_K60.3B46.22 GiB0.80 GiB47.70 GiB0.30 GiB17±37%
v6-Finch-14B-HFQ8_014.1B14.55 GiB32.41 GiB47.70 GiB0.30 GiB3±25%
MiniMax-M2.1-REAP-139B-A10BMoEI1-IQ2_S139B38.75 GiB8.23 GiB47.67 GiB0.33 GiB6±37%
m51Lab-MiniMax-M2.7-REAP-139B-A10BMoEI1-IQ2_S139B38.75 GiB8.23 GiB47.67 GiB0.33 GiB6±37%
KafkaLM-70B-German-V0.1Q4_069.0B36.20 GiB10.63 GiB47.65 GiB0.35 GiB3±25%
Qwen3-Coder-NextMoEQ4_K_S79.7B43.71 GiB3.19 GiB47.58 GiB0.42 GiB12±37%
Qwen3-Next-80B-A3B-ThinkingMoEQ4_K_S81.3B43.71 GiB3.19 GiB47.58 GiB0.42 GiB12±37%
Qwen3-Next-80B-A3B-InstructMoEQ4_K_S81.3B43.71 GiB3.19 GiB47.58 GiB0.42 GiB12±37%
GLM-4.5VMoEI1-Q2_K_S108B40.74 GiB6.11 GiB47.58 GiB0.42 GiB7±37%
CalmeRys-78B-Orpo-v0.1I1-IQ3_M78.0B35.33 GiB11.42 GiB47.58 GiB0.42 GiB3±25%
calme-2.3-rys-78bIQ3_M78.0B35.33 GiB11.42 GiB47.58 GiB0.42 GiB3±25%
Assistant_Pepe_70BIQ4_XS70.6B36.11 GiB10.63 GiB47.56 GiB0.44 GiB3±25%
Laguna-S-2.1MoEUD-IQ3_S118B45.10 GiB1.67 GiB47.49 GiB0.51 GiB13±37%
Step-3.5-Flash-REAP-121B-A11BI1-IQ2_S121B33.40 GiB13.30 GiB47.42 GiB0.58 GiB3±25%
deepseek-llm-67b-chatIQ4_XS67.4B34.00 GiB12.62 GiB47.42 GiB0.58 GiB3±25%
Devstral-2-123B-Instruct-2512IQ2_XS125B34.75 GiB11.69 GiB47.30 GiB0.70 GiB3±25%
Mistral-Medium-3.5-128BI1-IQ2_XS128B34.75 GiB11.69 GiB47.30 GiB0.70 GiB3±25%
XORTRON-NXTXPRTXXLI1-IQ2_XS128B34.75 GiB11.69 GiB47.30 GiB0.70 GiB3±25%
c4ai-command-r-plus-08-2024IQ3_XXS104B37.87 GiB8.50 GiB47.25 GiB0.75 GiB3±25%
ALIA-40b-fc-2606Q8_040.4B40.02 GiB6.38 GiB47.21 GiB0.79 GiB3±25%
ALIA-40b-instruct-2606Q8_040.4B40.02 GiB6.38 GiB47.21 GiB0.79 GiB3±25%
Llama-4-Scout-17B-16E-InstructMoEKV unresolvedQ2_K109B40.03 GiB6.38 GiB47.14 GiB0.86 GiB7±37%
deepseek-llm-67b-baseI1-IQ4_XS67.4B33.71 GiB12.62 GiB47.13 GiB0.87 GiB3±25%
openbuddy-deepseek-67b-v15.3-4kI1-IQ4_XS67.4B33.71 GiB12.62 GiB47.13 GiB0.87 GiB3±25%
Qwen3-42B-A3B-2507-Thinking-Abliterated-uncensored-TOTAL-RECALL-v2-Medium-MASTER-CODERMoEQ8_042.4B41.98 GiB4.45 GiB47.13 GiB0.87 GiB8±37%
DeepSeek-Coder-V2-Instruct-0724MoEIQ1_S236B44.14 GiB2.24 GiB47.12 GiB0.88 GiB12±37%
DeepSeek-V2.5MoEIQ1_S236B44.14 GiB2.24 GiB47.12 GiB0.88 GiB12±37%
DeepSeek-Coder-V2-InstructMoEIQ1_S236B44.14 GiB2.24 GiB47.12 GiB0.88 GiB12±37%
GLM-4.6VMoEIQ2_M108B40.26 GiB6.11 GiB47.10 GiB0.90 GiB7±37%
Maenad-70BIQ4_XS70.6B35.64 GiB10.63 GiB47.09 GiB0.91 GiB3±25%
DeepSeek-R1-Distill-Llama-70B-Uncensored-v2-Unbiased-ReasonerIQ4_XS70.6B35.64 GiB10.63 GiB47.09 GiB0.91 GiB3±25%
Strawberrylemonade-L3-70B-v1.2IQ4_XS70.6B35.64 GiB10.63 GiB47.09 GiB0.91 GiB3±25%
Llama-3.3_70_b_uncensored_continuedIQ4_XS70.6B35.64 GiB10.63 GiB47.09 GiB0.91 GiB3±25%
Llama-3.3-70B-Instruct-abliteratedIQ4_XS70.6B35.64 GiB10.63 GiB47.09 GiB0.91 GiB3±25%
grok-oss-Revenant-70BIQ4_XS70.6B35.64 GiB10.63 GiB47.09 GiB0.91 GiB3±25%
Hermes-4-70B-hereticIQ4_XS70.6B35.64 GiB10.63 GiB47.09 GiB0.91 GiB3±25%
Golem-70B-v1bIQ4_XS70.6B35.64 GiB10.63 GiB47.09 GiB0.91 GiB3±25%
DeepSeek-R1-Distill-Llama-70B-hereticIQ4_XS70.6B35.64 GiB10.63 GiB47.09 GiB0.91 GiB3±25%
Infinity-Instruct-7M-Gen-Llama3_1-70BIQ4_XS70.6B35.64 GiB10.63 GiB47.09 GiB0.91 GiB3±25%
GLM-4.5-Air-REAP-82B-A12BMoEQ3_K_L81.9B40.10 GiB6.11 GiB46.93 GiB1.07 GiB7±37%
Huihui-Qwen3-Coder-Next-abliteratedMoEQ4_K_L79.7B45.44 GiB0.80 GiB46.92 GiB1.08 GiB18±37%
Apertus-70B-Instruct-2509IQ4_XS70.6B35.36 GiB10.63 GiB46.87 GiB1.13 GiB3±25%
NVIDIA-Nemotron-3-Super-120B-A12B-BF16MoEIQ1_S124B43.19 GiB2.92 GiB46.81 GiB1.19 GiB11±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?
2030 of 2118 indexed open-weight models fit a AMD Ryzen AI Max 385 (Radeon 8050S) at 65,536 context with q8_0 KV cache, the largest being Qwen3.5-88B at I1-Q4_0. 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.