Hi everyone,
I’m posting after trying everything I could on my side. I’m running EndeavourOS, and like many others, I’m stuck with the “Data Center” drivers (first 595, now 610) that the distro forces on boot.
My setup:
- GPU: RTX 2080 Ti (EVGA XC Ultra)
- CPU: i5-12600KF
- RAM: 32GB
- Motherboard: MSI Gaming
- AIO: Freezer 3
- Driver: 610 (forced by pacman -Syu, can't stay on 595)
The problem:
- Space Marine 2, Satisfactory, and other DX12 games crash (black screen, “stopped working”)
- The driver artificially limits power draw to 292W max – impossible to go above
- However, my card can physically handle 300–350W peaks (temps stay <60°C)
- Light games, emulators, Warframe: everything runs fine
What I’ve tried:
- nvidia-smi -pl 300 → rejected (max 292W)
- Persistence mode, power smoothing tweaks, VRAM cache cleaning, screen stabilization
- Switching from Wayland to X11
- Auto game detection script (based on GPU usage)
Nothing works. The driver keeps crashing on demanding games.
My theories (personal, but based on testing):
- NVIDIA deliberately limits Turing cards on Data Center drivers to push newer hardware (or to keep gaming on Windows)
- The 292W limit is purely software – my card has the connectors, cooling, and capacitors to handle 350W
- A set of hidden bottlenecks (timeout, power smoothing, DPM) cause the crashes, but they’re hardcoded or locked at runtime
- The ideal fix would be an interception layer (LD_PRELOAD) to lie to the driver about GPU state (VRAM, voltage, power spikes)
Questions for the community:
- Is there any way to force a power limit above 292W on the 595/610 drivers?
- Does downgrading to the 570 driver remove this limit?
- Has anyone managed to run Space Marine 2 on a 2080 Ti without crashes?
- Any leads on disabling power smoothing or increasing the timeout via modprobe or a patch?
It’s a shame. This card is still great for both gaming and AI. Artificially capping it at 292W is a waste. And forcing Data Center drivers on a rolling distro like EndeavourOS is, in my opinion, a questionable decision.
Thanks in advance for any insights.
You’ll increase your chances of getting an answer a lot by using English. And you don’t have to be worried about grammar or vocabulary, many here are not native English speakers.
Welcome to the community 
Thanks for the feedback, I’ll switch it to English.
Well I am not sure if you realize the nature of a rolling distro, it just updates its packages all the time. So there is always a chanche a update will break something, that is a risk you have to take when using a rolling distro. There is no forcing this or forcing that it is just the nature of a rolling distro that you have to take for granted . If that it is not what you want , you might want to search for a distro that doesn’t get updates all the time.
I understand your point of view, but the fact that the packages are available piques my curiosity. Since the kernel is Arch, I don’t understand why we have to force things so much to interact with it and are content with the crumbs offered by Nvidia. And I debugged the 610 driver that crashed from the recovery terminal. Nvidia drivers are just awful, and we need to make it stable with the right access.
Not entirely shure what you mean, but it is a given fact EOS is for the biggest part based on whatever is available on ARCH (including the AUR) and most of the updates are coming from ARCH.
Almost all packages on EndeavourOS are directly Archlinux repositories. Nvidia driver is still propritary stuff.
Nvidia decides about possible features and default set values.
Look here:
You can see a Dev maintaining these from cachyOS he changed some default options some times ago making it way easier to be used. But the driver and utils are basically an Nvidia product. You can also use NVK/nouveau/mesa fully open source.
But what even is your card? inxi -Gaz
I’m using a 2080 Ti, it’s a Turing. That’s what I thought too, but the problem is that it’s stuck on data center drivers. And when I try to change the settings, it gets stuck, and it’s locked onto these drivers, forcing them to boot. This risks damaging Nvidia and having to clean it up and reinstall it. It wouldn’t take much to stabilize it.
what does that mean?
what exactly?
btw. codeblock in your answer is empty, i do not see the inxi -Gaz output.
Unless I’m mistaken, I understood that these drivers were for data centers. I wanted to increase the power limit to 320W, but instead I got the error “Privileged power limit 320 is not a valid power limit which should be between 100W and 292W.” Even when forcing it, I got the error “Max frequencies to increase performance: The requested functionality has been deprecated.” And the timeout error returned “No such file or directory.” Looking at nvidia-smi, it seems perfectly feasible; the instability is caused by the voltage spikes. The driver panic has normal voltage spikes, which is why I’m stuck trying to stabilize it.
Je reviendrais sur endevor quand Nova sera stable, vue que les driver ne son une prioriter pour la communauter. Merci pour vos retour.
I’ll come back to Endevor when Nova is stable, since drivers aren’t a priority for the community. Thanks for your feedback.
That’s non-sense, the Nvidia drivers are not specific for data center usage.
And they are stable, that’s not any beta or something simple latest Nvidia provides.
Also I must confess .. may language barrier?
I do not fully understand what exactly is your issue anyway. Were do you try to set that limits?
I do have 2 Nvidia system 3060 (pcie) and 5070 (Framework 16) both do not have such issues.
And we do not have anything to do with development of Nvidia drivers nor arch packaging.