ROCm and PyTorch recognize the GPU, as in, `torch.cuda.is_available` returns true, but I haven't actually run any models yet.
The maximum I can select on my 32GB RAM laptop is 8GB. Sounds like there are desktop AM4 mainboards where you can go up to 64GB.
One interesting thing I have seen being reported by `glxinfo` is auxiliary memory, which in my case is another 12GB, for a total of 20GB reported total available memory GPU memory. Unclear if this could be used by ROCm/PyTorch.
~2016 I bought a Lenovo ThinkPad E465 sporting an AMD Carrizo APU specifically to take advantage of HSA. It seemed like the feature, or at least the toolchain to take advantage of it, never really materialized. I'm glad at least someone else remembers it.
https://gpuopen.com/ags-sdk-5-4-improves-handling-video-memo...
Which says: "For APUs, this distinction is important as all memory is shared memory, with an OS typically budgeting half of the remaining total memory for graphics after the operating system fulfils its functional needs. As a result, the traditional queries to Dedicated Video Memory in these platforms will only return the dedicated carveout – and often represent a fraction of what is actually available for graphics. Most of the available graphics budget will actually come in the form of shared memory which is carefully OS-managed for performance."
The implication seems to be that you can have an arbitrary amount of graphics RAM, which would be appealing for AI use cases, even though the GPU itself is relatively underpowered.
Still, the question remains open, how to precisely control APU/GPU memory allocation on Linux and what is the limitations?
The biggest issue that keeps me from recommending it to anyone is that it is a windows 11 machine. That is both its strongest and weakest point. Turn it on for the first time, windows setup on a tiny screen with no physical mouse keyboard. And then there is messing with game graphics settings to eek out the performance you want. SD slot is complete garbage. Most likely a hardware issue that will not be able to be fixed. It stops reading cards and in some cases, destroys the card. It is fairly easy to upgrade the internal SSD though. ASUS has also been pretty quiet on graphics driver updates.
As long as you are cool with that stuff, its an awesome machine. Steamdeck has more of a 'just works' kinda console feel and kinda outdated hardware wise. Lenovo Go images just leaked today.