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. 1984 of 2118 indexed models fit at 16K 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 1704vision language 176audio tts 21image 2video 16audio asr 39embedding 26

What fits at 16K context

largest quantization that fits, per model · 1984 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
deepseek-llm-67b-chatI1-IQ2_M67.4B21.53 GiB1.67 GiB24.00 GiB0.00 GiB6±25%
deepseek-llm-67b-baseI1-IQ2_M67.4B21.53 GiB1.67 GiB24.00 GiB0.00 GiB6±25%
openbuddy-deepseek-67b-v15.3-4kI1-IQ2_M67.4B21.53 GiB1.67 GiB24.00 GiB0.00 GiB6±25%
spoomplesmaxx-v2.1-30BI1-Q6_K28.9B22.06 GiB1.13 GiB23.99 GiB0.01 GiB6±25%
Huihui-granite-4.1-30b-abliteratedI1-Q6_K28.9B22.06 GiB1.13 GiB23.99 GiB0.01 GiB6±25%
granite-4.1-30b-hereticI1-Q6_K28.9B22.06 GiB1.13 GiB23.99 GiB0.01 GiB6±25%
granite-4.1-30bQ6_K28.9B22.06 GiB1.13 GiB23.99 GiB0.01 GiB6±25%
Qwen3-Coder-Next-REAMMoEI1-IQ3_XS60.3B23.19 GiB0.11 GiB23.99 GiB0.01 GiB37±37%
OpenBuddy-R1-0528-Distill-Qwen3-32B-Preview0-QATQ5_K_L32.8B22.07 GiB1.13 GiB23.99 GiB0.01 GiB6±25%
KAT-DevQ5_K_L32.8B22.07 GiB1.13 GiB23.98 GiB0.02 GiB6±25%
Qwen3-VL-32B-InstructQ5_K_L33.4B22.07 GiB1.13 GiB23.98 GiB0.02 GiB6±25%
DeepSWE-PreviewQ5_K_L32.8B22.07 GiB1.13 GiB23.98 GiB0.02 GiB6±25%
Qwen3.5-122B-A10B-hereticMoEI1-IQ1_S123B23.13 GiB0.11 GiB23.97 GiB0.03 GiB35±37%
Bielik-11B-v2.3-InstructQ8_011.2B22.34 GiB0.88 GiB23.95 GiB0.05 GiB6±25%
GLM-4.7-FlashMoEQ6_K31.2B23.00 GiB0.23 GiB23.94 GiB0.06 GiB26±37%
Nemotron-Cascade-2-30B-A3BMoEQ4_K_M31.6B23.03 GiB0.23 GiB23.94 GiB0.06 GiB29±37%
Yi-34B-200K-DARE-megamerge-v8I1-Q5_K_S34.4B22.08 GiB1.05 GiB23.92 GiB0.08 GiB6±25%
dolphin-2.9.1-yi-1.5-34b-hereticQ5_K_S34.4B22.08 GiB1.05 GiB23.92 GiB0.08 GiB6±25%
dolphin-2.9.1-yi-1.5-34bI1-Q5_K_S34.4B22.08 GiB1.05 GiB23.92 GiB0.08 GiB6±25%
OrionStar-Yi-34B-Chat-LlamaI1-Q5_K_S34.4B22.08 GiB1.05 GiB23.92 GiB0.08 GiB6±25%
Yi-34B-200K-LlamafiedI1-Q5_K_S34.4B22.08 GiB1.05 GiB23.92 GiB0.08 GiB6±25%
Yi-1.5-34BQ5_K_S34.4B22.08 GiB1.05 GiB23.92 GiB0.08 GiB6±25%
Nous-Hermes-2-Yi-34BQ5_034.4B22.08 GiB1.05 GiB23.92 GiB0.08 GiB6±25%
Merged-RP-Stew-V2-34BI1-Q5_K_S34.4B22.08 GiB1.05 GiB23.92 GiB0.08 GiB6±25%
Capybara-Tess-Yi-34B-200KQ5_034.4B22.08 GiB1.05 GiB23.92 GiB0.08 GiB6±25%
Nous-Capybara-limarpv3-34BQ5_034.4B22.08 GiB1.05 GiB23.92 GiB0.08 GiB6±25%
Nemotron-Labs-Audex-30B-A3BQ4_K_M32.0B23.16 GiB0.00 GiB23.90 GiB0.10 GiB6±25%
archangel_sft-kto_llama30bIQ4_XS32.5B16.28 GiB6.86 GiB23.90 GiB0.10 GiB6±25%
magnum-v2-32bQ5_K_L32.5B21.95 GiB1.13 GiB23.87 GiB0.13 GiB6±25%
GLM-4.7-Flash-DerestrictedMoEI1-Q6_K31.2B22.92 GiB0.23 GiB23.87 GiB0.13 GiB26±37%
Huihui-GLM-4.7-Flash-abliteratedMoEI1-Q6_K31.2B22.92 GiB0.23 GiB23.87 GiB0.13 GiB26±37%
GLM-4.7-Flash-Claude-Opus-4.5-High-Reasoning-DistillMoEQ6_K31.2B22.92 GiB0.23 GiB23.87 GiB0.13 GiB26±37%
CodeLlama-70b-Instruct-hfI1-IQ2_M69.0B21.64 GiB1.41 GiB23.87 GiB0.13 GiB6±25%
CodeLlama-70b-Python-hfI1-IQ2_M69.0B21.64 GiB1.41 GiB23.87 GiB0.13 GiB6±25%
Nous-Hermes-Llama2-70bI1-IQ2_M69.0B21.64 GiB1.41 GiB23.86 GiB0.14 GiB6±25%
Midnight-Miqu-70B-v1.5I1-IQ2_M69.0B21.64 GiB1.41 GiB23.86 GiB0.14 GiB6±25%
Huihui-Qwen3.5-35B-A3B-abliteratedMoEI1-Q5_K_M36.0B23.06 GiB0.09 GiB23.86 GiB0.14 GiB35±37%
Qwen3.5-35B-A3B-BaseMoEI1-Q5_K_M36.0B23.06 GiB0.09 GiB23.86 GiB0.14 GiB35±37%
Qwen3.5-35B-A3B-ultra-uncensored-hereticMoEQ5_K_M35.1B23.06 GiB0.09 GiB23.86 GiB0.14 GiB35±37%
Qwen3.5-35B-A3B-Claude-4.6-Opus-Reasoning-DistilledMoEI1-Q5_K_M36.0B23.06 GiB0.09 GiB23.86 GiB0.14 GiB35±37%
Qwen3.6-35B-A3B-uncensored-hereticMoEQ5_K_M35.1B23.06 GiB0.09 GiB23.86 GiB0.14 GiB35±37%
Ornith-1.0-35B-uncensored-hereticMoEQ5_K_M35.1B23.06 GiB0.09 GiB23.86 GiB0.14 GiB35±37%
Nex-N2-mini-ultra-uncensored-hereticMoEQ5_K_M35.1B23.06 GiB0.09 GiB23.86 GiB0.14 GiB35±37%
Huihui-GLM-4.7-Flash-abliterated-57BMoEI1-IQ3_S57.3B22.52 GiB0.59 GiB23.85 GiB0.15 GiB24±37%
Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16UD-Q5_K_S33.0B23.10 GiB0.00 GiB23.85 GiB0.15 GiB6±25%
Qwen3-Next-80B-A3B-InstructMoEUD-IQ1_M81.3B22.73 GiB0.42 GiB23.85 GiB0.15 GiB34±37%
WizardCoder-Python-34B-V1.0I1-Q5_K_M33.7B22.20 GiB0.84 GiB23.84 GiB0.16 GiB6±25%
Phind-CodeLlama-34B-Python-v1I1-Q5_K_M33.7B22.20 GiB0.84 GiB23.84 GiB0.16 GiB6±25%
Phind-CodeLlama-34B-v2I1-Q5_K_M33.7B22.20 GiB0.84 GiB23.84 GiB0.16 GiB6±25%
CodeLlama-34b-instruct-hfQ5_K_M33.7B22.20 GiB0.84 GiB23.84 GiB0.16 GiB6±25%
WizardLM-1.0-Uncensored-CodeLlama-34bQ5_K_M33.7B22.20 GiB0.84 GiB23.84 GiB0.16 GiB6±25%
granite-34b-code-base-8kI1-Q5_K_M33.7B23.05 GiB0.00 GiB23.83 GiB0.17 GiB6±25%
Carnice-Qwen3.6-MoE-35B-A3BMoEI1-Q5_K_M36.0B23.03 GiB0.09 GiB23.82 GiB0.18 GiB35±37%
Qwen35B-Agent-R2-AbliteratedMoEI1-Q5_K_M34.7B23.03 GiB0.09 GiB23.82 GiB0.18 GiB35±37%
Qwen3.6-35B-A3B-Claude-4.6-Opus-Reasoning-DistilledMoEI1-Q5_K_M36.0B23.03 GiB0.09 GiB23.82 GiB0.18 GiB35±37%
Darwin-35B-A3B-OpusMoEI1-Q5_K_M36.0B23.03 GiB0.09 GiB23.82 GiB0.18 GiB35±37%
Qwen35B-Agent-R2MoEI1-Q5_K_M34.7B23.03 GiB0.09 GiB23.82 GiB0.18 GiB35±37%
Carnice-MoE-35B-A3BMoEI1-Q5_K_M36.0B23.03 GiB0.09 GiB23.82 GiB0.18 GiB35±37%
spoomplesmaxx-flash-35B-A3MoEI1-Q5_K_M35.1B23.03 GiB0.09 GiB23.82 GiB0.18 GiB35±37%
Huihui-Qwen3.6-35B-A3B-Claude-4.7-Opus-abliteratedMoEI1-Q5_K_M36.0B23.03 GiB0.09 GiB23.82 GiB0.18 GiB35±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?
1984 of 2118 indexed open-weight models fit a AMD Ryzen AI Max 385 (Radeon 8050S) at 16,384 context with q4_0 KV cache, the largest being deepseek-llm-67b-chat at I1-IQ2_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.