I've always been an NVIDIA user, but today I'm trialing an AMD R9 390. FPS-wise the performance is similar to my 780 Ti if not slightly worse. On the other hand, asynchronous timewarp seems to be much more reliable, not missing frames, even under load.
Is this normal, and is the stability of async tw likely to improve on NVIDIA prior to CV1 launch?
Update: I'm actually seeing stellar performance on a GTX 970 as well. I'm wondering if there's an architectural difference between the Kepler and Maxwell series which accounts for this, as a 970 and 780 Ti are not very in raw performance numbers.
Now I have to switch back to my 780 Ti to see whether that's really the case or whether a stealth update cleaned up performance across the board
There should be no major issues (in terms of performance) with supported cards from either brands. I believe even the 780 Ti should be OK, though maybe there is some strange software configuration that is causing issues.
I was seeing judder when rotating my head. Verifying with the Oculus Debug tool, the compositor was missing frames. I'd see it occasionally just in Oculus Home, but quite often in my application. Keep in mind, my application renders on multiple threads, pushes high-resolution textures between them, and generally abuses GPUs.
I'm not sure whether it was an Oculus Home upgrade, an NVIDIA driver upgrade, or the switch from 780 Ti to 970, but the issue is now completely gone. Rarely if ever do I see a missed frame, even in my own application 🙂
With certainly this what brought things into compliance. Preemption is a crazy animal and a number of things can break it, I suspect you picked up a fix around hangs in presentation.
Please let me know if you run into further issues. We can analyze further and see if there's something new going on.
