The AMD AI Max 395 does include integrated graphics. It comes equipped with the Radeon 8060S iGPU, which offers respectable standalone performance and should not be overlooked.
Qwen3.5-27B dense AI LLM model architecture performs about as good as Qwen3.5-122B-A10B MoE AI LLM model architecture: huggingface.co/Qwen/Qwen3.5-122B-A10B#benchmark-results.
And there is Qwen3.6-27B, but not Qwen3.5-122B-A10B.
Fitting a 27B UD-Q4_K_XL quant (17.9 GB) (huggingface.co/unsloth/Qwen3.6-27B-MTP-GGUF) requires 4.5 times less memory and nowhere near 128 GB RAM. All you need is 32 GB RAM + 6-8 GB VRAM GPU (basically any gaming PC/laptop).
More here:
"Your own ChatGPT, offline: AI without the cloud on your laptop":
notebookcheck.net/Your-own-ChatGPT-offline-AI-without-the-cloud-on-your-laptop.1341872.0.html
-> notebookchat.com/index.php?topic=315954.0.
Quote from: SinoRocket on July 26, 2026, 13:55:52The AMD AI Max 395 does include integrated graphics. It comes equipped with the Radeon 8060S iGPU, which offers respectable standalone performance and should not be overlooked.
Funny you say that, because, while AMD has announced FSR 4 support for its Radeon, RDNA 3, desktop GPUs, it hasn't so for its Radeon, RDNA 3.5 APUs, like this very (expensive) Strix Halo Radeon 8060S iGPU/APU XD.
So performance-wise, this could be on par with steam machine, just being smaller (even though probably also lauder and more expensive)...
Quote from: Cheaper alternative on July 26, 2026, 19:02:00And there is Qwen3.6-27B, but not Qwen3.5-122B-A10B.
Correction: "but not Qwen3.6-122B-A10B"