Dankpods used 12 computers with different hardware to test the performance of 5 games in 1080p and 4K, comparing the average fps results of the games’ built in benchmarks to determine which OS ran the game better across the same hardware: Windows or Bazzite.
Some notes on methodolgy under this spoiler
Each game uses the same in game graphics settings in Windows and in Linux. The Linux distro used was Bazzite, using the version specific for the graphics card hardware fpr each individual machine. To be clear, this means that he installed the Bazzite version for (legacy) nVidia as appropriate.
Each bazzite install was fresh, no copying installs or swapping around a drive with it pre-installed. After install, it was updated using system update and rebooted, repeated until no updates remained.
Screenshots of some of Dankpods’s comments to this effect:
There are many comments under the youtube video pointing out that in many of the Linux runs, it was not actually using the correct driver, comments about the experience using other distros, and comments about various potential fixes and workarounds.
This misses the point. Dankpods intentionally tested this way, and used Bazzite, to try and show what this would be like for the average gamer schmuck without a ton of technical skill interested in switching to Linux. Out of box experience matters in this situation, even though it’s not quite fair to compare that between free opens source distros and an OS created by a megacorp. To the average end user, it won’t matter. They just want it to work.
Prepare to be upset. With this particular testing methodology, Linux doesn’t really win overall.
I’m interested to hear the community’s thoughts on this.
Dankpods
How do we all just have the same personally and follow the same YouTubers.
Anyway his car channel is good, too
I thought they were the same bloke when I heard it!
It looks like most of the Nvidia systems were not running the proprietary Nvidia drivers (580/590) but instead falling back to NVK or even LLVMpipe (CPU rendering). All of the tested Nvidia GPUs are supposed to run using the proprietary driver on Bazzite. So, assuming that he downloaded the correct images, Bazzite really screwed up here.
But, unfortunately for the video, this doesn’t really show the typical gaming experience on Linux, it just highlights a Bazzite bug (?).
That’s the issue though. When everything works it’s great. But it’s so easy to bungle something up (be it user error or bugs in distros).
I’m running a 4070. Performance is really nice. Modern games pinned at vsync speed of 144FPS. The next day I’m down to 0.2FPS. Stays like that for a few days. Reboots don’t help. Can’t find anything debugging or googleing. After a few days it’s back up.
Turns out, I run games through heroic installed via flatpak, and flatpak keeps its own copy of Nvidia drivers. That version needs to be perfectly in sync with the system driver version. So
dnf updatebreaks that and the games fall back to CPU rendering, andflatpak updateplus reboot fixes it again. Runningflatpak updatefirst followed bydnf updatemakes sure performance always sucks.Took me a very long time to figure that out, and I imagine someone without an IT background might never figure that out.
Agreed. I think the main takeaway from the video is that it’s still hard to set up Nvidia GPUs on Linux, even on Bazzite :(
I love flatpak for how easy it makes it to install apps on almost any distro, but I also hate the spokes that it puts in the wheels. Drivers are ugly (that’s true for containers in general) and I also often stumble over file system permissions issues :(
I think it does show the typical experience. Its not a good test of linux systems, but it does show an accurate experience for a normal person trying out Linux on their existing hardware. These are exactly the problems unexperienced will run into. You see it all the time in help posts on linux communities. How we fix it, I’m not sure, but we shouldn’t deny that this is what its like for most people casually testing it out.
If that was the intent, then I think this was a very bad way to show that. A much better way would’ve been showing that it didn’t work on system X and resolving it (e.g. with some external help). Instead he just showed a large number of invalid/irrelevant benchmarks. This now leaves people thinking that Linux has a massive performance deficit instead of an issue with the driver installation. I would like to see a follow-up to address the driver issues and explain what went wrong, s.t. we can actually learn something from this.
I would also hope that the typical experience is that it works out of the box, especially on a distro like Bazzite, but that’s besides the point.
Yeah, I hope he does a part 2 where he gets some help to get it working, summarizes how, and re-runs the benchmarks.
How would I see if a game takes the correct driver? I‘m new to Bazzite.
In the video, it was sometimes showing the driver version at the end of the benchmarks (e.g. in Horizon Zero Dawn). If it says llvmpipe or NVK there, it’s not using the proprietary Nvidia driver.
But, if you want to check if your Nvidia GPU is detected correctly, you can run
nvidia-smion the terminal and if it shows you the installed GPU and driver, then it’s using the proprietary driver. Most desktop environments also have a “System Information” / “About this system” screen to show this information in the GUI.MangoHud may be able to show the driver too, but I’ve never used that myself.
Sounds like useful post-install steps that should be made clear for the nVidia release of Bazzite ;)
Yes, they could even show a list of detected GPUs, the driver used, and some status indication (e.g. warning if NVK is used).
I think this is why *there’s a bug reporting system.
I understand why he did it that way, but I completely disagree with the assessment that was all an end user should do after installing a fresh is on a new PC. Whenever I start up a fresh install of Windows or Linux I always make sure everything is working properly before I really do anything else. Is my GPU working properly, is my printer working, can I hear sound, etc. I just assumed that was standard.
Most people turn the computer on, set up an account and start using it.
The testing methodology is BAD, though. This is Youtuber science at its worst.
I understand the point of drawing attention to the OOBE and agree it’s valid, but that’s the kind of discussion best reserved for an article (or sure, video) specifically on the topic. This is an hour long video of collated bad data. It’s worthless, who wants to watch that unless you’re simply looking for validation or ragebait =/
You said the methodology was bad, twice, bit you didn’t even bother explaining why.
This misses the point. Dankpods intentionally tested this way, and used Bazzite, to try and show what this would be like for the average gamer schmuck without a ton of technical skill interested in switching to Linux. Out of box experience matters in this situation, even though it’s not quite fair to compare that between free opens source distros and an OS created by a megacorp. To the average end user, it won’t matter. They just want it to work.
From the post itself and when I specifically referenced the OOBE in my own post. You need to read and make certain connections yourself, I can’t connect every point for every one.
While there is a point to be made about the performance directly out of box, this assumes that the user would not eventually seek to resolve the issues to improve performance. While there is a valid point to be made on the overall experience and the difficulty of correcting these issues, comparing the performance between sets of correct and incorrect drivers does not provide valuable data. It just underlines the OOBE point over and over again, I don’t need to watch an hour long video for that point to be made.
Clear enough?
Why would a user want to go out of their way to resolve issues if there’s an alternative that just works. I mean that’s the line Linux fans always use “it just works” yet here is a clear example of it not working and now it’s apparently a user issue. If the users can’t configure the software correctly it’s inferior to the software that they can.
If you’d respond to that comment it would have made a lot more sense. What if the sorting of the comments put that post at the bottom and yours at the top? You can’t just assume people would connect the dots when you respond to the wrong thing.
Year of the Linux Desktop ™ confirmed - Cromulent!
After watching the first round, I’m glad I’m running two of the 3 cards that doninate on Linux.
My only question, which I feel wasn’t clearly explained in the video, is whether he did any extra work on the Windows machines. He explains his “fresh install” mythology for the Linux tests, I don’t recall him explicitly saying that he did the same for the Windows machines.
I’d be surprised if Windows actually ships with the newest drivers for the newer cards. For apples to apples, either run both OSes out of the box, or get proper driver’s both.
Fun video regardless!
Add on top of this modding, piracy and trouble shooting. I still think at this moment for a dedicated gaming machine (glorified console) windows is the better option. I believe everyone who says they can get equal to better performance and do all the aforementioned things on Linux; I don’t believe anyone who says everyone can do it or that it’s just as easy as doing on Windows.
Piracy works same as windows, you can download the same releases as you would for windows and install em via Heroic.
Modding depends on the game. Most any games which don’t rely on a mod loader (and many which do) are the same as windows. I did have issues with the doom rtx gzdoom version though.
