AMD · unified x86

AMD Ryzen AI Max+ 395 (Radeon 8060S)

AMD Ryzen AI Max+ 395 (Radeon 8060S) has 128 GB of VRAM at 256 GB/s — about 89.28 GiB usable after driver and compositor overhead. 2042 of 2118 indexed models fit at 128K context with f16 KV. Note only 96 GB of its 128 GB is allocatable to the GPU.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
128 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 1749vision language 189audio tts 21image 2audio asr 39embedding 26video 16

What fits at 128K context

largest quantization that fits, per model · 2042 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
MiniMax-M2MoEUD-IQ1_M229B64.01 GiB31.00 GiB95.70 GiB0.30 GiB2±37%
Step-3.5-Flash-REAP-121B-A11BI1-IQ3_XS121B45.92 GiB49.03 GiB95.68 GiB0.32 GiB2±25%
Mistral-Medium-3.5-128BI1-IQ3_S128B50.77 GiB44.00 GiB95.63 GiB0.37 GiB2±25%
XORTRON-NXTXPRTXXLI1-IQ3_S128B50.77 GiB44.00 GiB95.63 GiB0.37 GiB2±25%
MiMo-V2.5MoEKV unresolvedIQ2_XXS311B79.83 GiB15.00 GiB95.58 GiB0.42 GiB4±37%
Devstral-2-123B-Instruct-2512Q3_K_S125B50.63 GiB44.00 GiB95.49 GiB0.51 GiB2±25%
MiniMax-M2.1MoEUD-IQ1_M229B63.74 GiB31.00 GiB95.42 GiB0.58 GiB2±37%
MiniMax-M2.5MoEUD-IQ1_M229B63.74 GiB31.00 GiB95.42 GiB0.58 GiB2±37%
NVIDIA-Nemotron-3-Super-120B-A12B-BF16MoEUD-Q5_K_S124B83.56 GiB11.00 GiB95.25 GiB0.75 GiB4±37%
Llama-3.3-70B-InstructQ6_K_L70.6B54.39 GiB40.00 GiB95.21 GiB0.79 GiB2±25%
Anubis-70B-v1.2Q6_K_L70.6B54.39 GiB40.00 GiB95.21 GiB0.79 GiB2±25%
Tess-R1-Limerick-Llama-3.1-70BQ6_K_L70.6B54.39 GiB40.00 GiB95.21 GiB0.79 GiB2±25%
Infinity-Instruct-7M-Gen-Llama3_1-70BQ6_K_L70.6B54.39 GiB40.00 GiB95.21 GiB0.79 GiB2±25%
Athene-70BQ6_K_L70.6B54.39 GiB40.00 GiB95.21 GiB0.79 GiB2±25%
MiMo-V2-FlashMoEKV unresolvedIQ2_S310B79.34 GiB15.00 GiB95.08 GiB0.92 GiB4±37%
MiniMax-M2.7-BF16-ultra-uncensored-hereticMoEI1-IQ2_S229B63.36 GiB31.00 GiB95.05 GiB0.95 GiB2±37%
MiniMax-M2.7MoEIQ2_XS229B63.21 GiB31.00 GiB94.89 GiB1.11 GiB2±37%
Apertus-70B-Instruct-2509Q6_K70.6B53.95 GiB40.00 GiB94.83 GiB1.17 GiB2±25%
Meta-Llama-3-70B-InstructQ6_K70.6B53.92 GiB40.00 GiB94.75 GiB1.25 GiB2±25%
calme-2.4-llama3-70bQ6_K70.6B53.91 GiB40.00 GiB94.74 GiB1.26 GiB2±25%
calme-2.2-llama3-70bQ6_K70.6B53.91 GiB40.00 GiB94.74 GiB1.26 GiB2±25%
L3.3-70B-Magnum-v4-SEQ6_K70.6B53.91 GiB40.00 GiB94.74 GiB1.26 GiB2±25%
L3.3-Electra-R1-70bQ6_K70.6B53.91 GiB40.00 GiB94.74 GiB1.26 GiB2±25%
Latxa-Llama-3.1-70B-Instruct-v2I1-Q6_K70.6B53.91 GiB40.00 GiB94.74 GiB1.26 GiB2±25%
Llama-3.3_70_b_uncensored_continuedI1-Q6_K70.6B53.91 GiB40.00 GiB94.74 GiB1.26 GiB2±25%
grok-oss-Revenant-70BI1-Q6_K70.6B53.91 GiB40.00 GiB94.74 GiB1.26 GiB2±25%
Hermes-4-70BQ6_K70.6B53.91 GiB40.00 GiB94.74 GiB1.26 GiB2±25%
Llama-3.3-70B-Instruct-abliteratedQ6_K70.6B53.91 GiB40.00 GiB94.74 GiB1.26 GiB2±25%
Llama-3.1-70BQ6_K70.6B53.91 GiB40.00 GiB94.74 GiB1.26 GiB2±25%
Hermes-4-70B-hereticI1-Q6_K70.6B53.91 GiB40.00 GiB94.74 GiB1.26 GiB2±25%
Llama-3.1-Nemotron-70B-Instruct-HFQ6_K70.6B53.91 GiB40.00 GiB94.74 GiB1.26 GiB2±25%
L3.3-70B-Euryale-v2.3Q6_K70.6B53.91 GiB40.00 GiB94.74 GiB1.26 GiB2±25%
Rombos-LLM-70b-Llama-3.3Q6_K70.6B53.91 GiB40.00 GiB94.74 GiB1.26 GiB2±25%
llama-3-firefunction-v2Q6_K70.6B53.91 GiB40.00 GiB94.74 GiB1.26 GiB2±25%
DeepSeek-R1-Distill-Llama-70B-hereticI1-Q6_K70.6B53.91 GiB40.00 GiB94.74 GiB1.26 GiB2±25%
DeepSeek-R1-Distill-Llama-70BQ6_K70.6B53.91 GiB40.00 GiB94.74 GiB1.26 GiB2±25%
DeepSeek-R1-Distill-Llama-70B-abliteratedQ6_K70.6B53.91 GiB40.00 GiB94.74 GiB1.26 GiB2±25%
Legion-V2.1-LLaMa-70BQ6_K70.6B53.91 GiB40.00 GiB94.74 GiB1.26 GiB2±25%
Assistant_Pepe_70BI1-Q6_K70.6B53.91 GiB40.00 GiB94.74 GiB1.26 GiB2±25%
SEMIKONG-70BQ6_K70.6B53.91 GiB40.00 GiB94.74 GiB1.26 GiB2±25%
functionary-medium-v3.2KV unresolvedQ6_K70.6B53.91 GiB40.00 GiB94.74 GiB1.26 GiB2±25%
Llama-3.1-WhiteRabbitNeo-2-70BQ6_K70.6B53.91 GiB40.00 GiB94.74 GiB1.26 GiB2±25%
Hermes-3-Llama-3.1-70BQ6_K70.6B53.91 GiB40.00 GiB94.74 GiB1.26 GiB2±25%
L3.3-70B-Magnum-DiamondQ6_K70.6B53.91 GiB40.00 GiB94.74 GiB1.26 GiB2±25%
Meta-Llama-3-70B-Instruct-abliterated-v3.5Q6_K70.6B53.91 GiB40.00 GiB94.74 GiB1.26 GiB2±25%
Phi-3.5-MoE-instructMoEKV unresolvedF1641.9B78.00 GiB16.00 GiB94.71 GiB1.29 GiB3±37%
GLM-4.5-Air-REAP-82B-A12BMoEQ6_K81.9B70.98 GiB23.00 GiB94.71 GiB1.29 GiB2±37%
GLM-4.6VMoEQ5_K_S108B70.79 GiB23.00 GiB94.52 GiB1.48 GiB3±37%
Nex-N2-ProMoEIQ1_M397B90.02 GiB3.75 GiB94.52 GiB1.48 GiB7±37%
step-3.5-flashIQ2_XXS199B44.74 GiB49.03 GiB94.50 GiB1.50 GiB2±25%
GLM-4.5VMoEI1-Q5_K_S108B70.53 GiB23.00 GiB94.26 GiB1.74 GiB3±37%
Step-3.7-FlashIQ1_M201B44.41 GiB49.03 GiB94.17 GiB1.83 GiB2±25%
Behemoth-X-123B-v2Q3_K_S123B49.22 GiB44.00 GiB94.07 GiB1.93 GiB2±25%
Mistral-Large-Instruct-2411Q3_K_S123B49.22 GiB44.00 GiB94.07 GiB1.93 GiB2±25%
Llama-4-Scout-17B-16E-InstructMoEKV unresolvedQ5_K_S109B69.16 GiB24.00 GiB93.88 GiB2.12 GiB3±37%
CallerBF1632.8B61.04 GiB32.00 GiB93.83 GiB2.17 GiB2±25%
Dumpling-Qwen2.5-32BBF1632.8B61.04 GiB32.00 GiB93.83 GiB2.17 GiB2±25%
OpenThinker-32BF1632.8B61.04 GiB32.00 GiB93.83 GiB2.17 GiB2±25%
INTELLECT-2BF1632.8B61.04 GiB32.00 GiB93.83 GiB2.17 GiB2±25%
openhands-lm-32b-v0.1BF1632.8B61.04 GiB32.00 GiB93.83 GiB2.17 GiB2±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+ 395 (Radeon 8060S) run?
2042 of 2118 indexed open-weight models fit a AMD Ryzen AI Max+ 395 (Radeon 8060S) at 131,072 context with f16 KV cache, the largest being MiniMax-M2 at UD-IQ1_M. That covers text, vision-language, image, video and speech models.
How much usable memory does a AMD Ryzen AI Max+ 395 (Radeon 8060S) actually have?
Its nameplate is 128 GB, but about 89.28 GiB is available to a model once driver and compositor overhead is accounted for, and only 96 GB of the pool can be allocated to the GPU at all.
Is a AMD Ryzen AI Max+ 395 (Radeon 8060S) 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.