I'm testing the new client on mldstest. It seems faster! On my fast test machine (Ryzen 5950x) it'll complete a Parity DS1 WU in about 23 minutes (with no other WUs running). This client is a major change from the current production one, which is why it's taking so long.. The changes include:

* DS4 support
* Uses the BOINC wrapper instead of the native BOINC API (requires extensive backend changes)
* Jump to PyTorch v1.9 (up from v1.7)
* Statically compiled! (no more appimage or fuse requirement, and should support more Linux distros now, like CentOS 7)

I'm concerned that all these changes may introduce errors on hardware that I can't readily test. At the moment, only Linux/CPU is supported.. amd64 is out, the ARM builds are...progressing ... (raspberry pis are slow to build big complicated things like pytorch from scratch). And I'm still testing validation and the backend changes to work with the wrapper, but it's now open for testing and I encourage everyone on Linux to try it out... just expect some growing pains!

I'll also beable to pump out some DS4 test units too, hopefully.

Linux/amd64 only at the moment..even the arm builds are grumpy at me. Right now the priority for getting the new client to work is Linux/amd64 > Linux/arm > Windows/CPU > Linux/CUDA > Windows/CUDA > Linux/ROCm. I you're running Linux, please test!

Thanks for the short update! Looking forward to testing :)