It seems that the Windows driver already supports async reprojection for a long time, but until now it didn’t make it into the Linux version. Valve just recently released Half Life Alyx for Linux, and they do see some performance problem due to the lack of async reprojection on NV GPUs:
Specifically they need the high priority compute queue:
Is there a roadmap to get these things into the Linux driver sometime soon?
I did not find anything related in the release notes for the current 450 beta driver. Any new information on this?
I really hope we get support for async reprojection before the release of the 3000 series…
Ah so that explains it. I was wondering why it seemed to perform so poorly.
Anyone knows if there is a update regarding this?
What I currently do to optimize performance is actually disable interleaved reprojection.
I’m not sure if I understand how disabeling interleaved reprojection helps with optimizing the performance. Could you explain this a bit more?
If you’re targeting 90 FPS and you’re in situation where your hardware can only achieve 80 FPS then you’d still have to reproject every second frame, because frames simply take longer than the 11.1 ms / frame? Or how is this handled without interleaved reprojection?
NVidia please support it. We need this for good VR in Linux
I briefly got my hopes up when I saw ‘VK_EXT_global_priority’ (I think this is what’s missing on the Linux driver?) in the Vulkan Beta Driver notes, but it was only related to Windows 10 hardware scheduling…
I really hope that on the 28th AMD will show off something good…
Hi, NVIDIA. I am currently writing an article for a Linux gaming site and I would like to know when (or whether) you plan to have the high priority compute queue / asynchronous reprojection working. As the article should be up soon, if you don’t respond I assume you won’t, and I will stick to recommending AMD just like Valve did.
As an independent Linux software developer who’s working on a VR project and currently limited to NVIDIA hardware due to component availability in the marketplace, it’s frustrating to know there is no technical issue with regards to Linux and the implementation of a high priority compute queue. This feature is vital to support asynchronous reprojection and improving the VR experience on Linux. Implementation is just a matter of will. We already know the hardware supports it, and NVIDIA’s main competitor has shown the feature can be implemented on Linux.
As a follow up, I know many in the Linux community like me would appreciate a courtesy comment from NVIDIA as to whether or not they plan to implement the necessary support for asynchronous reprojection in their Linux driver. Perhaps it’s on their “to-do list” and they just haven’t gotten around to it due to higher priority items. Or perhaps they don’t think it’s worth their time. Either way, it would be nice to hear something official from NVIDIA instead of just speculating.
I waited more than a month for a response, but it seems it fell on dead ears. I published my article about VR on Linux and regretfully had to report that I got no response for this question. For reference, my article is on boilingsteam: https://boilingsteam.com/the-state-of-virtual-reality-on-linux/
BTW, due to that, I switched to the competition, I’m sorry to say.
Excellent article, patola! Sadly, I too will move over to AMD if NVIDIA refuses to acknowledge our concerns and support asynchronous reprojection, although I must wait until GPU availability improves. While the Linux market is small now, I do believe it is an important market and it will grow over time. I have found there are many technical advantages to having an open source operating environment.