NPU inclusion?

[KLiK]

Few words about future of tech & NPUs? What do you know? What do you think?

Is NPU going to be included into yoyo@h? What about some other project?

Will it be limited to CPU-NPU (like intel or AMD) or will it include some other NPU devices? Such as Googles Tensor & Coral TPU/NPU? Falcon cards?

Lets start the discussion here...

[Michael H.W. Weber]

At present, first of all there isn't much hardware available on the market. We (i.e. the Rechenkraft.net RNA World team) started "playing" a bit with Google's Coral board in combination with a PCIe-Card from ASUS which can harbor multiple of these cards - but Coral is limited to Tensorflow Lite which is bad for our purposes (i.e. employing machine learning models for biomolecular structure predictions and so on).
It looks like there is a true Tensorflow-capable HAT coming up for the Raspberry Pi 5 which is also much more powerful than the rather weak Coral board - that is something I am currently interested in but to the best of my knowledge it still can't be purschased in Germany.

Besides lack of affordable hardware availability of dedicated NPU cards there are of course a number of costly professional systems plus several affordable SBCs on the market. However, these mostly are designed for image recognition (chinese city mass surveillance applications) or lack proper documentation, software support and the like (SBCs). So, again nothing much to work with in practice.

Finally, in Yoyo@home at present there are no apps / projects that could take adavantage of NPUs / ML models.

Still, I believe that this will become a hot topic in the near future and might even have the potential to "replace" GPUs.

Michael.

[KLiK]

But what about those included in intel new gen CPU?

Link: https://www.intel.com/content/www/us/en ... ies-1.html

[Michael H.W. Weber]

But what about those included in intel new gen CPU?

Link: https://www.intel.com/content/www/us/en ... ies-1.html

...drivers for Windows only?

Michael.