VRAM Allocation Issues

Hi @amrits !
I made the video with bug repro. At 0:30 I open the map and fps drops to 10 and never recovers, GPU usage becomes constant 100% but temperature goes down. Opening the map is just the way to allocate a bit more VRAM. It happens without map as well when usage reaches 100%. Also if I open the map when VRAM usage is much lower then 100% everything is ok.
So, I think RTX 3050 mobile is the best GPU to investigate it because I can reproduce the same on almost every modern game.
Ping me if I can do something to help investigate this.

1 Like

That’s exactly how it is - the best ‘tester’ is Deathloop with everything on the max - under the RTX 4070 Ti after about 15 minutes of playing there is a hiccup as for the RTX 3050 … 100% with 12 GB VRAM. The same e.g. Cyberpunk 2077 on a mobile RTX 3060 with 6 GB VRAM - it doesn’t matter DLSS on Performace = 100% 6 GB already occupied in a moment on the highest settings and drop from e.g. 50 fps to 10 … under 520 drivers turning on DLSS on Performacne freeing a lot of VRAM - but on 525 there is a tragedy - it keeps VRAM busy all the time.

1 Like

Looks like I do have local repro now on below setup with game Cyberpunk 2077 with default game settings where VRAM consumption is around 6.2 GB out of 8 GB and FPS value ranges from 6-8 during the entire game play.
LENOVO_MT_4810_BU_Think_FM_ThinkPad P1 Gen 5 + Ubuntu 22.04.1 LTS + NVIDIA GeForce RTX 3070 Ti Laptop GPU + Driver 525.78.01
Please confirm if this can be considered as repro so that debugging can be done in right direction.

Can confirm sfjuocekr’s issues with Elite: Dangerous with Nvidia’s newer drivers. Using 530.41, visual artifacts have now rendered it unplayable.

Good news!
I don’t know if it is the same issue I talk about but at least it’s a nice point to start. I can repro it on my 3050 mobile only when VRAM usage reaches 100% (even for a short period). But maybe 100% usage is not the main reason of the issue but just a good situation for repro and issue can happen below 100% as well, just much less often.

I can confirm the rendering issues in Elite: Dangerous, triangles completely glitch out of position and flicker across the screen with geometry completely deformed, in-game cockpit HUD shows colored backgrounds. Downgrading to 525 fixes this. But I no longer have performance issues since using a 12 GB card with DXVK max memory set to 9 GB (the desktop already allocates 1.7-2 GB, DXVK has some slack overhead from its chunk allocator).

With my previous 6 GB card, it was almost impossible to play with a steady performance as getting in and out of the ship usually overshot VRAM allocations for a very short period of time, ending in high PCI-e bus utilization as render resources seem to have been allocated from system memory - even with DXVK max memory settings.

It looks like some games first allocate resources for a new scene (overshooting VRAM budget), and then immediately free the previous resources. It then looks like allocations are still within the size of VRAM when they really haven’t been for a very short moment.

Downgrading to the 525 drivers immediately makes a lot of these issues disappear. I’m in Arch Linux and just did a downgrade and edited pacman.conf to ignore nvidia package upgrades, including:

  • nvidia-dkms
  • nvidia-utils
  • lib32-nvidia-utils
  • lib32-opencl-nvidia
  • libxnvctrl
  • nvidia-settings
  • opencl-nvidia

Why are people hijacking this thread to report rendering issues? I first reported similar issues here:

It would be better if people replied there or made a new thread.

2 Likes

Highly related issue: Non-existent shared VRAM on NVIDIA Linux drivers

2 Likes

Hi All,
Please help to test with driver 535.104.05 or latest release drivers for high VRAM usage issue and share feedback.

1 Like

If a test with 545.23.06 is helpful: I cannot reproduce anymore the problem that I had with Star Citizen as described here, where intense memory pressure would cause the frame rate to drop to single-digit numbers. Now, even when the VRAM is used 100%, the frame rate remains stable.

Since SC has also changed in the meantime, it’s not a very clean test, so I guess it would be helpful if other people also share their experience with other games. In any case, thank you very much for all your hard work!

1 Like

Thanks @kerberizer
Shall wait for feedback from other users as well.

1 Like

Seems like performance issues related to high VRAM usage are at the very least greatly alleviated. Thanks for the fix.

1 Like

Great news !! Thanks for sharing the feedback.

Yep, it’s for sure much better than before, thanks for fixing it!

Nvidia GTX 970 4GB (545.29.02)
Arch Linux - KDE Plasma (5.27.9)

i can confirm that the above mentioned vram issues still persist on 535.154.05

sysconfig:
asus x13 - 5900hs, 3050ti 4GB, 16GB
ubuntu 23.10
liquorix 6.6.12-1
man. install 535.154.05

vram is being “eaten up” by any application using the dedicated gpu. even a virtual machine is doing that. assigned 256MB will increase the longer the machine is being in use.

as soon as vram reaches 3840-3880MB (when actually 4096MB should be available!) fps drops from 37-40s to 25-30s and stay there until game is being restartet. this applies to Any game.

these drivers can be called bloatware! because their bloating themselfs up for nothing.

tell me if you need any logs or so. thanks!

1 Like

Just as a followup, it seems like whatever “fix” is implemented breaks if you alt-tab out of a game.

NVIDIA really needs to fix this issue, makes any modern game unplayable. But on windows, VRAM management works perfectly because it relies on the system RAM too and this doesn’t happen with Linux NVIDIA drivers.

now the situation looks pretty clear to me. because the driver under linux cannot overlay its memory on the RAM, exactly these problems are created that we have been experiencing here for several driver versions. therefore, these problems are quite exotically variable and manifest themselves somewhat differently depending on the system and application, but with certain similarities.

you can almost call it a fraud that a company like nvidia will not be able to achieve operating system equality in 2024. we the end users rely on the data we have ->Before Purchasing Products<- and it says quite clearly - “linux driver available” and not “test software possible for installation”.

the excuse that there are “countless linux distributions and kernels” and therefore “you can’t address all the variables” simply doesn’t work anymore these days. a simple definition of the system requirements would be very easy to implement. for example, ubuntu lts with stock kernel - at least there everything should run smoothly.

please finally fix that. your stuff acts like toys made by kids!

I have vram problems with Palworld, reaching 100% of vram and the game’s fps dropping sharply. At the beginning of the game I can get more than 100fps, but when I start using the map’s fast travel, it’s as if the vram can’t clear the rest of the map loaded.

SO: Gentoo Linux
Graphics:
Device-1: NVIDIA GA104 [GeForce RTX 3060 Ti] driver: nvidia v: 535.154.05
Device-2: AMD Cezanne [Radeon Vega Series / Radeon Mobile Series]
driver: amdgpu v: kernel
Device-3: Logitech Webcam C270 driver: snd-usb-audio,uvcvideo type: USB
Display: x11 server: X.org v: 1.21.1.11 with: Xwayland v: 23.2.4 driver:
X: loaded: amdgpu,nvidia unloaded: modesetting dri: radeonsi gpu: amdgpu
resolution: 1920x1080
API: EGL v: 1.5 drivers: nvidia,radeonsi,swrast
platforms: x11,surfaceless,device
API: OpenGL v: 4.6 vendor: amd mesa v: 23.3.1 renderer: AMD Radeon
Graphics (radeonsi renoir LLVM 17.0.6 DRM 3.49 6.1.67-gentoo)
API: Vulkan v: 1.3.268 drivers: radv,nvidia surfaces: xcb,xlib

1 Like