[mindcrash]

> "The game changer is the unified 128 GB memory. That is the path Apple took years ago. Instead of separate memory for the CPU and GPU"

Not really a game changer as AMD already has this (but is quite terrible at marketing it). This is the reason why you can run pretty big AI models on reasonable speeds on embedded GPUs on the Ryzen AI SoCs and get even better performance on the Ryzen AI Max+ SoCs.

As a example I have a HX370 based laptop with 128G RAM and due to these specs can instruct the included firmware to share up to 96G (or approximately 66% of available system memory) with the embedded GPU, which gives quite some headroom to experiment with local models as said before.

The included GPU with 6144 CUDA cores isn't really big news either to be honest, as the DGX Spark already has them too. Likely because they simply integrated the same (or very similar) GPU on this SoC. In fact the whole thing looks eerily similar to the DGX Spark SoC, so the only new thing pretty much seems to be the support for Windows.