So missing this as yet another end user!
We are many. This issue invalidates VR on GeForce GPUs. Nvidia, we need a roadmap when this is going to be resolved.
The lacking of async reprojection is making many VR applications in Linux unfeasible, even with a 3090. How is this not important to NVidia/Valve, why is nobody working on this, giving feedback or sharing a roadmap for this. With all the effort poured into making Linux gaming actually work in the last years, I can’t see no reason why this has been a major issue for a year. Maybe that’s because there’s zero communication towards NVidia’s customers?
Valve cannot have any influence/interference on Nvidia driver as it is closed source.
Valve can do such things on open source drivers though, which they already do on Radv by having multiple devs working on it/funding various other areas too.
So this is strictly an Nvidia issue, Valve can’t fix it for them.
Exactly, but I would hope Valve would use its might to keep nagging them about it and inform their customers (us) about updates they receive from NVidia teams. Right?
Valve did everything they can, including calling them out publicly with Tweet’s shared above. (Dan Ginsburg is a Valve employee)
Everything is up to Nvidia really, Valve can’t force them more to fix it.
Valve: We did everything we can do with Nvidia.
Linux users: What exactly?
Valve: Our employee tweeted them.
Surely it is up to Nvidia to proceed with implementation, because drivers are closed. But saying that Valve did everything can do is exaggeration. Valve is critical partner to keep Nvidia GPU sales going, so they should have direct influence on Nvidia when it comes to API support. Issues like this should not hang in the forum thread, but be part of the driver release schedule with developers assigned.
The advantage to having an open-source driver is clear. If NVIDIA did not have the time and resources to implement the feature, someone from the open-source community would have done the work to add and test the needed functionality. In such a scenario, NVIDIA would win and all the Linux users would win. It would be a “win-win” solution, which is what most business people are taught to find. NVIDIA’s main competitor has implemented the feature and has an open-source driver for Linux. In terms of competition, this is a feature deficiency which NVIDIA needs to rectify. The VR market is only going to grow, and it’s going to grow on all platforms, not just Microsoft Windows.
Fellow Linux VR users:
It’s been almost a full year since Dan Ginsburg tweeted about the lack of a “high priority compute queue” in the Linux NVIDIA driver. The absence of this feature prevents SteamVR from supporting asynchronous reprojection on Linux NVIDIA systems. This results in a sub-optimal experience compared to similarly equipped AMD RADV systems, where the feature is available.
Of course this feature is available in the Windows NVIDIA driver. Unfortunately, the Linux driver has not maintained parity with its Windows cousin. Despite our concerns, we in the Linux community have received no response from NVIDIA.
As we contemplate this “radio-silence” from NVIDIA, the current GPU shortage and what our purchasing decisions will be once the shortage is over, I propose a toast to the forgotten OS. Grab your favorite beverage and raise it high.
To Linux. Linux is everywhere. It’s used for high-profile scientific, military, educational and even entertainment purposes. It’s even now on Mars, thanks to NASA and the Ingenuity helicopter.
And it’s used for virtual reality, too. If only the experience could be that much better when using an NVIDIA GPU.
1 year later and there has still not been any official word from nvidia on this issue.
I would appreciate an explanation why my expensive GPU is delivering a subpar VR experience on linux compared to a lower specced GPU from a competitor.
Any news on this subject?
Does Nvidia Vulkan driver supports the high priority scheduling mechanism needed to implement “async reprojection” on Linux yet?
I just installed the 465 driver on Ubuntu. The driver has some new Vulkan extensions according to the release notes, but those extensions didn’t look like a high priority compute queue.
Does anyone on this list know the name of the specific identifier of the extension they need to support? Maybe if we knew the technical name we could somehow ask for it specifically. Not sure how to get such a request in front of an NVIDIA developer’s eyeballs. I have no idea if they even read these forums or not.
To their credit, NVIDIA has been fixing bugs in the Linux driver, which is good. I just wish we could get some transparency on the asynchronous reprojection support.
Edit: Is it “VK_EXT_global_priority” which we need?
Edit #2: Apparently it is, according to the following link:
Edit #3: I’m running Ubuntu 20.04 with NVIDIA driver 465.27
Here’s what Steam’s vrcompositor log thinks of my system:
grep -i async ~/.steam/steam/logs/vrcompositor.txt
Wed Jun 02 2021 23:14:59.662721 - Attempting to enable async support…
Wed Jun 02 2021 23:14:59.662728 - Async support unavailable on Nvidia platforms.
But Vulkan seems to think the driver has the right extension:
vulkaninfo | grep -i global_priority
VK_EXT_global_priority : extension revision 2
And I saw something online about the vrcompositor needing “cap_sys_nice” enabled, which it is:
~/.steam/steam/steamapps/common/SteamVR/bin/linux64/vrcompositor-launcher = cap_sys_nice+eip
So does the string “Async support unavailable on Nvidia platforms” get printed to the log because SteamVR looks at the driver and assumes that if it is from NVIDIA it won’t work, or is SteamVR trying to start asynchronous reprojection, failing and then prints the message because it notices the driver is from NVIDIA? Either is possible I suppose.
Do you know if the VK_EXT_global_priority feature was missing in previous drivers?
460.67 also returns;
VK_EXT_global_priority : extension revision 2
Async support unavailable on Nvidia platforms.
So I don’t see change compared to your output.
I suspect the Linux NVIDIA driver is claiming it supports VK_EXT_global_priority, but it doesn’t actually support VK_EXT_global_priority. I could see that happening if some commits were pushed over to the Linux side, but not all the commits necessary to implement the feature.
Of course we won’t know anything until some NVIDIA engineer or manager actually notices this forum exists, reads these messages and then bothers to make a statement or take action. I’m not holding my breath on that one.
Async reprojection will be supported in our upcoming 470 driver series.
Great news! Thank you :)
Thank you Damien for informing us about NVIDIA’s plans to support async reprojection for Linux in a future update. I was wondering if anyone from NVIDIA was reading this forum. I’m very happy to see this issue is finally getting some attention!
Finally here’s the great news. Will be a game changer for us Linux VR peeps.
I’ve been testing with the NVIDIA 470.42.01 beta driver manually installed on Ubuntu 20.04. Overall it’s a great improvement. Turning my head no longer creates a jerky/janky experience in X-Plane while running in VR, which was my chief complaint.
I have noticed crashes in the beta version of Steam VR on three different games. I would say the experience is stable until it bugs out, and then VR just blanks out the headset display, but it doesn’t take the system with it so… yea? Not sure if this is a Steam VR issue or a NVIDIA driver issue or some kind of interaction between the two.
If any other Linux users would like to try the beta, here’s a link to NVIDIA’s advanced driver search, which is how I found it:
Direct link to the beta at this point in time: