AMD · unified x86

AMD Ryzen AI Max 390 (Radeon 8050S)

AMD Ryzen AI Max 390 (Radeon 8050S) has 32 GB of VRAM at 256 GB/s — about 22.32 GiB usable after driver and compositor overhead. 1956 of 2118 indexed models fit at 32K 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 1678vision language 174audio tts 21audio asr 39image 2video 16embedding 26

What fits at 32K context

largest quantization that fits, per model · 1956 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Qwen3-Coder-NextMoEUD-IQ2_XXS79.7B21.71 GiB1.59 GiB23.99 GiB0.01 GiB24±37%
Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTPIQ2_M27.8B22.16 GiB1.06 GiB23.99 GiB0.01 GiB6±25%
OpenBuddy-R1-0528-Distill-Qwen3-32B-Preview0-QATQ4_K_L32.8B18.94 GiB4.25 GiB23.99 GiB0.01 GiB6±25%
KAT-DevQ4_K_L32.8B18.94 GiB4.25 GiB23.98 GiB0.02 GiB6±25%
Qwen3-VL-32B-InstructQ4_K_L33.4B18.94 GiB4.25 GiB23.98 GiB0.02 GiB6±25%
DeepSWE-PreviewQ4_K_L32.8B18.94 GiB4.25 GiB23.98 GiB0.02 GiB6±25%
gemma-3-12b-it-abliterated-v2F1611.8B21.92 GiB1.31 GiB23.98 GiB0.02 GiB6±25%
gemma-3-12b-it-abliteratedBF1612.2B21.92 GiB1.31 GiB23.98 GiB0.02 GiB6±25%
gemma-3-12b-itBF1612.2B21.92 GiB1.31 GiB23.98 GiB0.02 GiB6±25%
14BQ5_114.2B9.95 GiB13.28 GiB23.98 GiB0.02 GiB6±25%
Salience-1.5-ProMoEQ5_K_S36.0B22.94 GiB0.33 GiB23.98 GiB0.02 GiB32±37%
Qwable-v1MoEQ5_K_S36.0B22.94 GiB0.33 GiB23.98 GiB0.02 GiB32±37%
T-SearchMoEQ5_K_S36.0B22.94 GiB0.33 GiB23.98 GiB0.02 GiB32±37%
codellama-13b-oasst-sft-v10Q6_K13.0B9.95 GiB13.28 GiB23.97 GiB0.03 GiB6±25%
chronos-hermes-13b-v2Q6_K13.0B9.95 GiB13.28 GiB23.97 GiB0.03 GiB6±25%
WhiteRabbitNeo-13B-v1Q6_K13.0B9.95 GiB13.28 GiB23.97 GiB0.03 GiB6±25%
CodeLlama-13b-Instruct-hfQ6_K13.0B9.95 GiB13.28 GiB23.97 GiB0.03 GiB6±25%
Orca-2-13b-Alpaca-UncensoredI1-Q6_K13.0B9.95 GiB13.28 GiB23.97 GiB0.03 GiB6±25%
WizardLM-13B-UncensoredI1-Q6_K13.0B9.95 GiB13.28 GiB23.97 GiB0.03 GiB6±25%
WizardCoder-Python-13B-V1.0I1-Q6_K13.0B9.95 GiB13.28 GiB23.97 GiB0.03 GiB6±25%
Guanaco-13B-UncensoredI1-Q6_K13.0B9.95 GiB13.28 GiB23.97 GiB0.03 GiB6±25%
Llama-2-13b-chat-hfQ6_K13.0B9.95 GiB13.28 GiB23.97 GiB0.03 GiB6±25%
Wizard-Vicuna-13B-UncensoredQ6_K13.0B9.95 GiB13.28 GiB23.97 GiB0.03 GiB6±25%
WizardLM-13b-V1.0-UncensoredQ6_K13.0B9.95 GiB13.28 GiB23.97 GiB0.03 GiB6±25%
MythoMax-L2-13bQ6_K13.0B9.95 GiB13.28 GiB23.97 GiB0.03 GiB6±25%
WizardLM-1.0-Uncensored-Llama2-13bQ6_K13.0B9.95 GiB13.28 GiB23.97 GiB0.03 GiB6±25%
speechless-llama2-hermes-orca-platypus-wizardlm-13bQ6_K13.0B9.95 GiB13.28 GiB23.97 GiB0.03 GiB6±25%
mythalion-13bQ6_K13.0B9.95 GiB13.28 GiB23.97 GiB0.03 GiB6±25%
Qwen3.6-35B-A3B-REAM-160-ru-agentMoEQ8_023.6B22.92 GiB0.33 GiB23.96 GiB0.04 GiB28±37%
Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-PreservedMoEQ5_K_S35.1B22.91 GiB0.33 GiB23.95 GiB0.05 GiB32±37%
Qwen3.5-35B-A3B-uncensored-heretic-v2-Native-MTP-PreservedMoEQ5_K_S35.1B22.91 GiB0.33 GiB23.94 GiB0.06 GiB32±37%
magnum-v2-32bQ4_K_L32.5B18.88 GiB4.25 GiB23.93 GiB0.07 GiB6±25%
Qwen3.6-35B-A3B-NVFP4-FastMoENVFP421.8B22.89 GiB0.33 GiB23.93 GiB0.07 GiB32±37%
Maenad-70BI1-IQ2_XXS70.6B17.79 GiB5.31 GiB23.92 GiB0.08 GiB6±25%
DeepSeek-R1-Distill-Llama-70B-Uncensored-v2-Unbiased-ReasonerI1-IQ2_XXS70.6B17.79 GiB5.31 GiB23.92 GiB0.08 GiB6±25%
Rombos-LLM-70b-Llama-3.3I1-IQ2_XXS70.6B17.79 GiB5.31 GiB23.92 GiB0.08 GiB6±25%
L3.3-Electra-R1-70bI1-IQ2_XXS70.6B17.79 GiB5.31 GiB23.92 GiB0.08 GiB6±25%
L3.3-70B-Magnum-v4-SEIQ2_XXS70.6B17.79 GiB5.31 GiB23.92 GiB0.08 GiB6±25%
Latxa-Llama-3.1-70B-Instruct-v2I1-IQ2_XXS70.6B17.79 GiB5.31 GiB23.92 GiB0.08 GiB6±25%
Llama-3.3_70_b_uncensored_continuedI1-IQ2_XXS70.6B17.79 GiB5.31 GiB23.92 GiB0.08 GiB6±25%
Llama-3.3-70B-Instruct-abliteratedI1-IQ2_XXS70.6B17.79 GiB5.31 GiB23.92 GiB0.08 GiB6±25%
grok-oss-Revenant-70BI1-IQ2_XXS70.6B17.79 GiB5.31 GiB23.92 GiB0.08 GiB6±25%
Llama-3.1-Nemotron-70B-Instruct-HFI1-IQ2_XXS70.6B17.79 GiB5.31 GiB23.92 GiB0.08 GiB6±25%
L3.3-70B-Euryale-v2.3I1-IQ2_XXS70.6B17.79 GiB5.31 GiB23.92 GiB0.08 GiB6±25%
Hermes-3-Llama-3.1-70BIQ2_XXS70.6B17.79 GiB5.31 GiB23.92 GiB0.08 GiB6±25%
Hermes-4-70B-hereticI1-IQ2_XXS70.6B17.79 GiB5.31 GiB23.92 GiB0.08 GiB6±25%
Llama-3.3-70B-InstructIQ2_XXS70.6B17.79 GiB5.31 GiB23.92 GiB0.08 GiB6±25%
Llama-3.1-70BIQ2_XXS70.6B17.79 GiB5.31 GiB23.92 GiB0.08 GiB6±25%
Anubis-70B-v1.2IQ2_XXS70.6B17.79 GiB5.31 GiB23.92 GiB0.08 GiB6±25%
Hermes-4-70BIQ2_XXS70.6B17.79 GiB5.31 GiB23.92 GiB0.08 GiB6±25%
Golem-70B-v1bI1-IQ2_XXS70.6B17.79 GiB5.31 GiB23.92 GiB0.08 GiB6±25%
DeepSeek-R1-Distill-Llama-70B-abliteratedI1-IQ2_XXS70.6B17.79 GiB5.31 GiB23.92 GiB0.08 GiB6±25%
DeepSeek-R1-Distill-Llama-70B-hereticI1-IQ2_XXS70.6B17.79 GiB5.31 GiB23.92 GiB0.08 GiB6±25%
DeepSeek-R1-Distill-Llama-70BIQ2_XXS70.6B17.79 GiB5.31 GiB23.92 GiB0.08 GiB6±25%
Legion-V2.1-LLaMa-70BI1-IQ2_XXS70.6B17.79 GiB5.31 GiB23.92 GiB0.08 GiB6±25%
Assistant_Pepe_70BI1-IQ2_XXS70.6B17.79 GiB5.31 GiB23.92 GiB0.08 GiB6±25%
Tess-R1-Limerick-Llama-3.1-70BIQ2_XXS70.6B17.79 GiB5.31 GiB23.92 GiB0.08 GiB6±25%
SEMIKONG-70BIQ2_XXS70.6B17.79 GiB5.31 GiB23.92 GiB0.08 GiB6±25%
functionary-medium-v3.2KV unresolvedIQ2_XXS70.6B17.79 GiB5.31 GiB23.92 GiB0.08 GiB6±25%
Llama-3.1-WhiteRabbitNeo-2-70BIQ2_XXS70.6B17.79 GiB5.31 GiB23.92 GiB0.08 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 390 (Radeon 8050S) run?
1956 of 2118 indexed open-weight models fit a AMD Ryzen AI Max 390 (Radeon 8050S) at 32,768 context with q8_0 KV cache, the largest being Qwen3-Coder-Next at UD-IQ2_XXS. That covers text, vision-language, image, video and speech models.
How much usable memory does a AMD Ryzen AI Max 390 (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 390 (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.