[Various/all Distros] Numerous Performance & Rendering Issues on 390.25
And I can also confirm the same for Linux Mint 18.3 (--no-sandbox flag resolves Chrome/Chromium/Opera issues).
And I can also confirm the same for Linux Mint 18.3 (--no-sandbox flag resolves Chrome/Chromium/Opera issues).

#91
Posted 03/10/2018 10:46 PM   
Hold up folks - lets actually be clear... --no-sandbox is a workaround for Chrome. It *does not* fix the actual problem. Don't go saying that workarounds resolve the issue as there are many other programs that have issues with this driver version that exist. tl;dr - don't act like Windows cancer users.
Hold up folks - lets actually be clear... --no-sandbox is a workaround for Chrome. It *does not* fix the actual problem.

Don't go saying that workarounds resolve the issue as there are many other programs that have issues with this driver version that exist.

tl;dr - don't act like Windows cancer users.

#92
Posted 03/11/2018 12:41 AM   
What are the other issues introduced by 390.25?
What are the other issues introduced by 390.25?

#93
Posted 03/11/2018 08:42 AM   
I have similar stutters in games (though most are WINE). Chromium stutters when an image is loaded, games stutter when resources are loaded (e.g. just before a new character model is rendered on the screen). F1 2017 (linux native) if it loads models that haven't been loaded before it stutters slightly. After doing a lap it's fine because all the assets are presumably cached. The first lap of any race stutters slightly at various points (though less noticeable than wine games) World of Warcraft (wine) - stutters don't last as long and it's not as obvious it's the same cause, but because things are rendered at far distances it likely is. 0.1 seconds is stutters Starcraft 2 it's easy to reproduce:select a unit of a type you haven't selected before (for maximum effect, while moving the camera) and the whole game stutters for up to half a second. Freelancer (old but I've been playing it recently) runs at 60ps but whenever a new ship is about to be rendered the game freezes for 0.2-0.5s and then the ship is rendered on the screen. (Note it's not just when the ship is rendered on screen, but presumably the game loads assets for other ships within a radius of the player, it's when a ship enters this radius). I'll record a video of the issue and upload it.
I have similar stutters in games (though most are WINE). Chromium stutters when an image is loaded, games stutter when resources are loaded (e.g. just before a new character model is rendered on the screen).


F1 2017 (linux native) if it loads models that haven't been loaded before it stutters slightly. After doing a lap it's fine because all the assets are presumably cached. The first lap of any race stutters slightly at various points (though less noticeable than wine games)

World of Warcraft (wine) - stutters don't last as long and it's not as obvious it's the same cause, but because things are rendered at far distances it likely is. 0.1 seconds is stutters

Starcraft 2 it's easy to reproduce:select a unit of a type you haven't selected before (for maximum effect, while moving the camera) and the whole game stutters for up to half a second.

Freelancer (old but I've been playing it recently) runs at 60ps but whenever a new ship is about to be rendered the game freezes for 0.2-0.5s and then the ship is rendered on the screen. (Note it's not just when the ship is rendered on screen, but presumably the game loads assets for other ships within a radius of the player, it's when a ship enters this radius).


I'll record a video of the issue and upload it.

#94
Posted 03/11/2018 03:26 PM   
"What are the other issues introduced by 390.25?" Permanent 14% GPU load when using CompositionPipeline and generally worse performance were what I was seeing. As well as black tearing across the screen.
"What are the other issues introduced by 390.25?"

Permanent 14% GPU load when using CompositionPipeline and generally worse performance were what I was seeing. As well as black tearing across the screen.

#95
Posted 03/11/2018 03:41 PM   
Video of Freelancer here.: https://youtu.be/beutRgtyvas (make sure you have youtube set to 60fps or it'll be less obvious--youtube doesn't show all resolutions right away, 60fps may not appear until an hour or so after this post). I chose freelancer as it's old and the fps never dips below 60. Sorry for the manic camera, it's easier to demonstrate it when there is constant movement. Typically while recording the stutters didn't seem quite so bad, but you can see what happens at about 10s. There are at least 3 stutters in that 30 second clip, makes playing games annoying as hell. edit: I should mention the game is running from a Samsung 960 evo nVME drive, it's not a I/O bottleneck. I've tried the following: __GL_THREADED_OPTIMIZATIONS on/off. Changing the nvme's scheduler Forcing performance mode rather than adaptive in the nvidia driver Changing the cpu govener to performance taskset to force the game to a specific 2/4/8 cores Changing the bios to use local mode for memory (threadripper numa/uma feature) While running the game there are no CPU spikes when the stutter occurs (unless they are too short for monitoring software to see, I was using ksysguard)
Video of Freelancer here.: https://youtu.be/beutRgtyvas (make sure you have youtube set to 60fps or it'll be less obvious--youtube doesn't show all resolutions right away, 60fps may not appear until an hour or so after this post). I chose freelancer as it's old and the fps never dips below 60.

Sorry for the manic camera, it's easier to demonstrate it when there is constant movement. Typically while recording the stutters didn't seem quite so bad, but you can see what happens at about 10s. There are at least 3 stutters in that 30 second clip, makes playing games annoying as hell.


edit: I should mention the game is running from a Samsung 960 evo nVME drive, it's not a I/O bottleneck.

I've tried the following:


__GL_THREADED_OPTIMIZATIONS on/off.
Changing the nvme's scheduler
Forcing performance mode rather than adaptive in the nvidia driver
Changing the cpu govener to performance
taskset to force the game to a specific 2/4/8 cores
Changing the bios to use local mode for memory (threadripper numa/uma feature)

While running the game there are no CPU spikes when the stutter occurs (unless they are too short for monitoring software to see, I was using ksysguard)

#96
Posted 03/11/2018 04:14 PM   
Has anyone tried the new [url=http://www.nvidia.com/download/driverResults.aspx/131853/en-us]390.42[/url]? Do they have the same performance issues? The changelog says: [quote]Fixed a regression, introduced in 390.12, that caused occasional flicker when using the X driver's composition pipeline, for example when using screen transformations like rotation, or the "ForceCompositionPipeline" or "ForceFullCompositionPipeline" options."[/quote]
Has anyone tried the new 390.42? Do they have the same performance issues?

The changelog says:

Fixed a regression, introduced in 390.12, that caused occasional flicker when using the X driver's composition pipeline, for example when using screen transformations like rotation, or the "ForceCompositionPipeline" or "ForceFullCompositionPipeline" options."

#97
Posted 03/12/2018 07:37 PM   
[quote=""]Has anyone tried the new [url=http://www.nvidia.com/download/driverResults.aspx/131853/en-us]390.42[/url]? Do they have the same performance issues? The changelog says: [quote]Fixed a regression, introduced in 390.12, that caused occasional flicker when using the X driver's composition pipeline, for example when using screen transformations like rotation, or the "ForceCompositionPipeline" or "ForceFullCompositionPipeline" options."[/quote][/quote] Nope, definitely not fixed. For the record, I don't use the composition pipeline.
said:Has anyone tried the new 390.42? Do they have the same performance issues?

The changelog says:

Fixed a regression, introduced in 390.12, that caused occasional flicker when using the X driver's composition pipeline, for example when using screen transformations like rotation, or the "ForceCompositionPipeline" or "ForceFullCompositionPipeline" options."

Nope, definitely not fixed. For the record, I don't use the composition pipeline.

#98
Posted 03/12/2018 07:53 PM   
I can confirm that this new 390.42 DOES NOT FIX the chromium or other performance issues. Just to set the record the problems exists with the default setting of no composition pipeline as I don't use these settings. Now we probably have to wait another month before there is even a chance of a fix happening. They will have to release a new driver sooner though because Xorg 1.20 is just about finished and Linux 4.16 is arriving at the beginning of April. Hopefully they have already reproduced it at least, it took me less than 2 minutes to notice the problem in Chrome after I installed 390.25 so it shouldn't be that hard to at least see the problem.
I can confirm that this new 390.42 DOES NOT FIX the chromium or other performance issues.

Just to set the record the problems exists with the default setting of no composition pipeline as I don't use these settings. Now we probably have to wait another month before there is even a chance of a fix happening.

They will have to release a new driver sooner though because Xorg 1.20 is just about finished and Linux 4.16 is arriving at the beginning of April.

Hopefully they have already reproduced it at least, it took me less than 2 minutes to notice the problem in Chrome after I installed 390.25 so it shouldn't be that hard to at least see the problem.

Dell Precision M4700
Intel QM77 chipset
Core i7 3940XM (with Intel HD4000 graphics)
Nvidia Quadro K2000M (Geforce GT650m equivalent)
16 GB DDR3 1600 ram
Windows 10 Pro 1709
Arch Linux with KDE Plasma

#99
Posted 03/12/2018 09:13 PM   
Same problem here (25-30fps in with kde, on a geforce 1080ti).. After downgrading from 390-25 to 384.111, I get a constant 60fps.
Same problem here (25-30fps in with kde, on a geforce 1080ti)..

After downgrading from 390-25 to 384.111, I get a constant 60fps.

Posted 03/12/2018 09:43 PM   
I know. It's obvious that something is causing everything to use a slow path on download to the GPU. Should be pretty easy to bisect.
I know. It's obvious that something is causing everything to use a slow path on download to the GPU. Should be pretty easy to bisect.

Posted 03/12/2018 09:45 PM   
This thread is already 7 pages, so I am sure that I don't need to lend any information that has already been given. I definitely have performance issues exactly as described by other users. Vivaldi, Chromium, and Chrome are all affected among other applications. I just wanted to lend another voice so Nvidia knows this is a big issue affecting lots of people. Hardware Macbook Pro 11,2 Geforce 750M 390.25 OS: Antergos/(Arch) Kernel: 4.15.7
This thread is already 7 pages, so I am sure that I don't need to lend any information that has already been given. I definitely have performance issues exactly as described by other users. Vivaldi, Chromium, and Chrome are all affected among other applications. I just wanted to lend another voice so Nvidia knows this is a big issue affecting lots of people.

Hardware
Macbook Pro 11,2
Geforce 750M
390.25

OS: Antergos/(Arch)
Kernel: 4.15.7

Posted 03/12/2018 11:04 PM   
I'm still getting this like a flood in my journalctl whenever I open or exit application that uses GPU. This is even with nvidia-persistenced enabled. [code]kernel: nvidia-modeset: Allocated GPU:0 (GPU-b909bbcf-2700-dce3-c390-11d51818c438) @ PCI:0000:01:00.0 kernel: nvidia-modeset: Freed GPU:0 (GPU-b909bbcf-2700-dce3-c390-11d51818c438) @ PCI:0000:01:00.0[/code]
I'm still getting this like a flood in my journalctl whenever I open or exit application that uses GPU. This is even with nvidia-persistenced enabled.

kernel: nvidia-modeset: Allocated GPU:0 (GPU-b909bbcf-2700-dce3-c390-11d51818c438) @ PCI:0000:01:00.0
kernel: nvidia-modeset: Freed GPU:0 (GPU-b909bbcf-2700-dce3-c390-11d51818c438) @ PCI:0000:01:00.0

Posted 03/12/2018 11:29 PM   
[quote=""][quote=""]Has anyone tried the new [url=http://www.nvidia.com/download/driverResults.aspx/131853/en-us]390.42[/url]? Do they have the same performance issues? The changelog says: [quote]Fixed a regression, introduced in 390.12, that caused occasional flicker when using the X driver's composition pipeline, for example when using screen transformations like rotation, or the "ForceCompositionPipeline" or "ForceFullCompositionPipeline" options."[/quote][/quote] Nope, definitely not fixed. For the record, I don't use the composition pipeline.[/quote] Same here. 390.42 does not fix performance issues for me either. And for the record, I can't even enable Composition Pipeline in xorg.conf file because Xorg won't start with Force( or Full)CompositionPopeline enabled.
said:
said:Has anyone tried the new 390.42? Do they have the same performance issues?

The changelog says:

Fixed a regression, introduced in 390.12, that caused occasional flicker when using the X driver's composition pipeline, for example when using screen transformations like rotation, or the "ForceCompositionPipeline" or "ForceFullCompositionPipeline" options."

Nope, definitely not fixed. For the record, I don't use the composition pipeline.


Same here. 390.42 does not fix performance issues for me either.

And for the record, I can't even enable Composition Pipeline in xorg.conf file because Xorg won't start with Force( or Full)CompositionPopeline enabled.

Posted 03/13/2018 02:01 AM   
[quote=""]I know. It's obvious that something is causing everything to use a slow path on download to the GPU. Should be pretty easy to bisect.[/quote] I think the devs know, although I doubt the fixes are so obvious: they were trying to do something different for a reason, I imagine. 390.25 will not be a Hall of Fame long-term release, for sure. But downgrading seems like a good fix unless you are on a really new kernel. Look on the bright side: we get all the fun of Wayland (eg no screensharing), for free!
said:I know. It's obvious that something is causing everything to use a slow path on download to the GPU. Should be pretty easy to bisect.

I think the devs know, although I doubt the fixes are so obvious: they were trying to do something different for a reason, I imagine. 390.25 will not be a Hall of Fame long-term release, for sure. But downgrading seems like a good fix unless you are on a really new kernel.

Look on the bright side: we get all the fun of Wayland (eg no screensharing), for free!

Posted 03/13/2018 02:46 AM   
Scroll To Top

Add Reply