Hello,

I have a workstation that runs a Quadro P2000. I have Slackware64 -current installed running KDE4. It is up to date with the -current branch. I've installed the latest NVIDIA drivers from here. https://www.nvidia.com/Download/driv...x/163238/en-us

For the most part, everything works ok. But, I'm noticing some performance issues on certain applications, and Kwin. Kwin for example will stutter mildly. And when resizing Firefox's window for example lags quite bad. Firefox also seems to have some performance issues. On most sites it's ok, but Reddit for example is quite laggy. The Firefox interface is laggy too. I have tried no HW acceleration (Basic), OpenGL, and Webrender in Firefox, but they're all about the same.

Glxinfo shows NVIDIA is the provider. Games like Minecraft run fine, and utilize the GPU well. Running things like Folding@Home can fully utilize the GPU without issue. The laggyness happens without running folding though.

I also get tearing if I don't enable full composite pipeline in the NVIDIA settings. I read though this can hurt performance, and introduce input lag. I have tried setting the Kwin env variables __GL_YIELD to USLEEP, and also KWIN_TRIPLE_BUFFER to 1, but tearing still occurs. I tried enabling triple buffering in xorg.conf, but tearing occurs. So, the only way to fix that seems to be enabling full composite pipeline. Is there any other fix for this?

Other things I've tried include setting, "options nvidia NVreg_UsePageAttributeTable=1" in /etc/modprobe.d/nvidia.conf with no performance change. Forcing nvidia-drm=1 causes the modesetting to fail, and the boot output to freeze till X starts. Do I need to build in the nvidia kernel modules to my initrd?

What other setting can I adjust to help improve performance?

Any help would be appreciated!

Maybe narrow down the list of issues, it appears there are at least 4 different problems that you describe.
Each of those fixes and workarounds you've mentioned is hardware specific and driver version specific, and yet you don't even mention the driver version.
IMO It's better to revert all those fixes and workarounds until you find the source of the problem, otherwise you can't be sure they are not causing even more problems.
I could suggest a few things, though.. Check first if your "folding" hasn't melted the card. If it's bent, get it fixed or replaced.
I'd check also if the same type of lag happens in other environments/distributions, or for example XFCE where tearing can sometimes simply be avoided by forcing sync_to_vblank on.
The (sync_to_vblank in xfce settings editor (xfwm channel)) setting works fine on some cards and is often used to avoid tearing, but not all chips react the same to this setting.
If you find your problem's specific to KWIN, install latest supported version of KWIN to see if that'll fix it.
KDE4 is not supported upstream anymore, so I guess that would mean QT5, frameworks, plasma upgrade.

There's also a "few" other threads about these issues on the forum already:
https://www.linuxquestions.org/questions/tags/nvidia/

And a thread where latest drivers are discussed, even though they aren't distributed in Slackware:
https://www.linuxquestions.org/quest...se-4175608923/

I use ForceCompositionPipeline instead of ForceFullCompositionPipeline. I don't feel the performance hit with it and it fixes the tearing I experienced with my GTX 1070. Also I almost forgot. I set I did this a few years ago because I didn't like the new one and how it affected the composition pipeleine.

Maybe you can find answers here: https://forums.developer.nvidia.com/...hics/linux/148

1 members found this post helpful.

Greetz
I'm not very familiar how much overlap there is between gaming oriented models and Quadro (the most recent Quadro I have is in a T61P Thinkpad) but I'm pretty sure some performance profile tweaking is available within nvidia-settings, from the simple performance preference to the clock increasing profiles available once coolbits is employed.... all under the Performance tab.

Another consideration that should be examined is cooling. Potent GPUs throttle down when they get hotter and some profiles actually begin with GPUI fan(s) Off! I like silent operation but the "operation" is the key element for me. I set my fans to run a lot faster than stock profiles, and I blow th4e cruft out of the heatsink fins and fans regularly.

Thank you all who replied! In terms of thermals for this GPU, it actually remains very cool! Even when under 100% utilization while folding, the temps never raise above 77c-79c. The NVIDIA settings say the slowdown temp isn't till 101c. Under full load the GPU fan runs at about 67%. As I type this, the GPU is running at only 44c, and the fan is at 52%.

I tried RadicalDreamer's suggestion, and it has worked wonders! I reverted all the other changes I made, then just enabled the "ForceCompositionPipeline" setting, and set "Option "UseNvKmsCompositionPipeline" "false"" in my xorg.conf. Everything is much more responsive! I was also able to re-enable the "Allow Flipping" option without it causing graphical issues in Firefox now. Kwin animations are also a lot smoother! Thank you RadicalDreamer for suggesting this! Overall, I'm happy with the performance now!

1 members found this post helpful.

Are you guys saying the "UseNvKmsCompositionPipeline" "false" option has an effect even on a non KMS system? I assumed since OP downloaded the driver from nVidia that it was a runlevel 3 .run install, No?

You are correct, I did install the NVIDIA driver, version 450.66, via the .run file while in runlevel 3. I'm not entirely sure why this option would work, but it certainly does seem to make a noticeable difference. I just tested by commenting it out and rebooting. With "UseNvKmsCompositionPipeline" "false" commented out the Firefox interface lag returns. However, after removing the comment and rebooting, it resolves that issue. Kwin also seems to stutter less with "UseNvKmsCompositionPipeline" "false" set in xorg.conf. So, it does seem to have an effect.

1 members found this post helpful.

aplattner said, "The display composition pipeline went through a major overhaul in release 390. As a side effect, it is now re-compositing the screen on every vblank. That’s likely the cause of the GPU load you’re seeing. We’re working on optimizing it to reduce the overhead (tracking bug 2085439) but in the meantime, you can force the driver to use the old composition pipeline by setting"
https://forums.developer.nvidia.com/...0-25/57733/154

I noticed a change for the worst in one driver update and found that was the solution to the performance degradation. It is sad that it is still needed but I'm glad I can use it.

I know this is probably OT, but what is everyone's take on NVIDIA's new hotness, the RTX 30xx series? I can hardly believe that everyone and their dog is willing to burn their hard-earned (or stimulus!) money on a card barely more performant than the previous generation. Or it could just be my imagination.

I think it is bang for the buck. The 3070 and 3080 cost almost half as much as the previous line of Ray Tracing boards and have similar performance to their substantially more expensive predecessors. I have a GTX 1070 Ti which, when overclocked slightly beats a 1080 Ti but not only can't match any of the 2000 series can't do Ray Tracing and that's going to be really big soon. I'll probably wait for a year but the 3000 series is on my radar.

I saw the Quake 2 RTX demo awhile back, and that was some impressive stuff, but that was (I think) a calculated risk by NVIDIA to ramp up the hype for the ray-tracing technology. It's been around for decades (At least 4 [or a little over!], by my count), but only now has it come into its own, becoming the next generation of gaming goodness. We'll see soon enough, but I'm certainly not getting all that excited over it just yet.

Happy Slacking!

1 members found this post helpful.

New consoles are coming and we will receive ports of games made for them. I'm holding off on getting more hardware until I see the new game requirements. I'll probably get a 5070-6070 with a 7nm-5nm Intel cpu or 3-5nm AMD processor if/when they come out in a few years.