It doesn’t really make sense for Zynthian, the open source synths and plugins which drive Zynthian wouldn’t use the 128 core GPU.
I think that Modart (Pianoteq) guys could take huge profit of using such a beast for running their models. Same for any true-modelling synthesizer, specially if modelling analogue instruments.
Also, it could be nice for calculating big convolution matrixes, useful for implementing some effects. You could simulate real-life spaces, rooms, cabinets, lesslies, etc.
Anyway, it’s a long way for this and currently no software used by zynthian can take profit of this beast 
But it could allow 8 stream of hi-res graphics to sync along with a zynth . . . .
That has got to be a fairly interesting visualization at some point!
The Jetson boards are miracles. A colleague used the TX2 for a project and I loved it and the incredible speed it reached running a customized TensorFlow to do object detection on 16 720p video streams at the same time, and encoding a 4K dashboard with the 16 video inputs scaled and with recognition output overlayed…
Never thought to made a synth with it… 
Think about those A.I. controlled PureData generative patch possibilities though!
when will zynthian support neuroevolutionary synthesis? can you run PyTorch as a VST layer in Zynthian? I really need this for a project.
VST is not supported and won’t be in the near future.
No problem! With Jetson we simply could train a model with plugins that currently have LV2 and VST versions and it would learn to write ports 
Until the above model is ready, we only can run PyTorch as LV2 plugin
Also TensorFlow & Caffe will be available in such a way 
