Strix Halo (Ryzen AI Max+ 395, Radeon 8060S) for local LLMs

32 measured results from 9 builds, each linked to its source.

Can it run? Gemma 4 26B-A4B on Strix Halo (Radeon 8060S) Gemma 4 31B on Strix Halo (Radeon 8060S) Kolibri-1 on Strix Halo (Radeon 8060S) Qwen3.6 27B on Strix Halo (Radeon 8060S) Qwen3.6 35B-A3B on Strix Halo (Radeon 8060S) Qwen3.8 27B on Strix Halo (Radeon 8060S) gpt-oss 120B on Strix Halo (Radeon 8060S)

All builds with the Strix Halo (Ryzen AI Max+ 395, Radeon 8060S)

All measured results

Hide speculative-decoding results ⚡ (MTP, DFlash, DSpark and similar draft-and-verify methods give much higher tok/s on predictable text and are not comparable 1:1 with plain decode)

All Strix Halo (Ryzen AI Max+ 395, Radeon 8060S) builds: see also all GPUs · best LLMs by memory size · download the dataset.

FAQ

How fast is the Strix Halo (Ryzen AI Max+ 395, Radeon 8060S) for local LLMs?

The fastest plain-decode result we have is 88.14 tok/s on Llama 3.2 3B (Q4_K_M, llama.cpp Vulkan (iGPU, llama-bench tg128)) in Framework Desktop (AMD Strix Halo) mainboard. Speed depends heavily on the model, quant and engine, so compare like with like in the table. RigBase lists 32 measured results from 9 builds.

How much memory does the Strix Halo (Ryzen AI Max+ 395, Radeon 8060S) have?

unified memory: 96–128 GB in our builds. Memory bandwidth: 256 GB/s (memory bandwidth as listed on llamabench).

Which models have been measured on the Strix Halo (Ryzen AI Max+ 395, Radeon 8060S)?

DeepSeek R1 Distill 70B, DeepSeek R1 Distill 8B, Gemma 4 26B-A4B, Gemma 4 31B, Kolibri-1, Llama 3.1 70B, Llama 3.2 3B, Qwen3.6 27B, Qwen3.6 35B-A3B, Qwen3.8 27B, Qwen3.8 Flash-Next, gpt-oss 120B.

Where do these numbers come from?

Every row links to a build page with its original public source (benchmark logs, READMEs, forum posts). RigBase does not re-run the benchmarks. Results marked ⚡ use speculative decoding and are not comparable 1:1 with plain decode.