Is 2025 the Year of the Linux Desktop - Part II

AUTHOR:Arthur B.
TESTED BY:Arthur B.
Published:

Introduction

Welcome back! In the last article I discussed the testing methodology, test bench set up, and my qualitative thoughts on setting up Ubuntu for GPU testing. In this article I will explore how the tested cards performed on Ubuntu vs Windows at 1080p, 1440p, and 1080p raytraced. 

A reminder that the following cards were tested: 

  • AMD RX 7700 XT 12GB

  • AMD RX 9070 16GB

  • Intel Arc B580 12GB

  • NVIDIA RTX 4060 Ti 16GB

  • NVIDIA RTX 5060 Ti 16GB

I’ll also present the results of a few ad hoc tests I performed, including an upscaling performance test and a distro war – okay more like a distro skirmish. By the end of this article, I will have answered the question on everyone’s minds: Is 2025 the year of the Linux desktop?

AMD RX 7700 XT

The results were off to a promising start in native rasterization. At 1080p the 7700 XT on Ubuntu 24.04 was about 4.6% slower than on Windows 11, and at 1440p it was only about 4.2% slower. Ray tracing is where Windows took the biggest lead over Ubuntu 24.04, averaging a 53.1% increase across our two games. Overall, these results are far from discouraging for Linux gamers rocking AMD GPUs in their systems, as the measured performance difference in native rasterization shouldn’t be too noticeable during gameplay. You will probably notice the performance hit if you like to enable raytracing, but hopefully this will continue to improve as new driver and proton versions get released.

1 of 3

AMD RX 9070

As mentioned in the last article, the 9070 doesn’t work with the kernel that ships with Ubuntu 24.04. There are ways to switch the kernel version in the currently installed OS, but this is an advanced task that requires tinkering in the terminal. Thus, in keeping with my Noob-centric approach, I chose to switch to Ubuntu 25.04 since this is one of the simplest ways for a new user to get access to a more recent kernel version. As expected, the newer version solved the issues and the 9070 was immediately detected.

The performance of this card was quite impressive considering it was released a mere 2 months before I started testing. This bodes well for the future if the support from the open source community continues. In Black Myth: Wukong the 9070’s performance between Windows and Ubuntu was nearly identical at 1080p and 1440p. Ubuntu even managed a slight victory over Windows 11 in Strange Brigade at both resolutions.  At 1080p, Ubuntu was 8% slower than Windows on average in native rasterization, and 1440p was a similar story with an 8.5% performance decrease in Ubuntu compared to Windows. Raytracing was a weak point for Ubuntu with Windows performing 110% better in Black Myth: Wukong and 70% better in Cyberpunk 2077. 

1 of 3

Intel Arc B580

In the previous article, I mentioned some of the strange driver issues encountered with the B580 on Ubuntu 24.04. As a reminder, the gaming performance on the B580 was quite disappointing compared to its Windows counterpart. At 1080p, in native rasterization, the B580 was 30.4% slower in Ubuntu compared to Windows, and this trend continued at 1440p with Ubuntu being 29.4% slower than Windows.

1 of 2

Not wanting to give up entirely on the B580, I decided to give the games that failed in Ubuntu 24.04 a second shot on Ubuntu 25.04. Unfortunately, Shadow of the Tomb Raider through Proton and Forza Horizon 5 still failed, but the visual issues were fixed in both Cyberpunk 2077 and Black Myth: Wukong. 

The results on this new driver were much more encouraging. In 1080p raster, Ubuntu lagged by 8% behind Windows in Black Myth: Wukong and was 27.4% behind in Cyberpunk 2077. Unfortunately, I was unable to get raytracing working on the B580. Enabling raytracing in Cyberpunk 2077 with the in-game settings seemed to work, but upon closer inspection it was clear that the game was still running with native rasterization. There may have been a way to enable raytracing on the B580, and Intel’s driver website definitely implied that there should have been, but I was unable to get raytracing running. Overall, these results had me hopeful for the future of Intel’s Battlemage architecture on Linux, but further improvements are still needed for the experience to be competitive  with AMD.

1 of 2

NVIDIA RTX 4060 Ti

In native rasterization at 1080p the 4060 Ti was 17.7% slower on Ubuntu 24.04 than on Windows 11 across the tested games. Increasing the resolution to 1440p, led to similar results with Ubuntu being about 17.5% slower. This performance delta was quite a bit higher than on the AMD cards, which just goes to show what can be achieved when hardware vendors provide open source developers sufficient support. Hopefully NVIDIA’s announced support for open source driver development will yield improvements in the near future. The silver lining for NVIDIA was that the 4060 Ti had a smaller gap in ray tracing performance between Windows and Ubuntu than what was seen on the RX 9070 (but not the RX 770XT). At 1080p with ray tracing enabled, Windows was 57.6% faster than Ubuntu across Cyberpunk 2077 and Black Myth Wukong.

1 of 3

NVIDIA RTX 5060 Ti

At 1080p in native rasterization the 5060 Ti was 30% faster in Windows 11 than on Ubuntu 24.04. The results were similar at 1440p in native rasterization with Windows being 29.6% faster on average in the game suite than Ubuntu. Raytracing performance once again seemed to be a weak point for Ubuntu. Windows was 73.3% faster than Ubuntu in Black Myth: Wukong, and 27.9% faster in Cyberpunk 2077. 

1 of 3

Impact of Upscaling Technologies

In the grim darkness of the far future, there is only AI slop. Whether you are gaming on Windows or Linux, there is no escaping the AI powered upscaling technologies that have flooded the market.

In order to see if these technologies took a performance hit from Proton, I tested the performance of DLSS, FSR, and XeSS by running Cyberpunk 2077 at 1440p Ultra with the respective technologies set to “Quality”. I wanted to present the upscaling technologies in their best possible light, so I tested DLSS on the NVIDIA RTX 4060 Ti, FSR on the AMD RX 9070, and XeSS on the Intel Arc B580. All tests were performed on Ubuntu 25.04 and Windows 11 24H2 for comparison. Note that I would have liked to test DLSS on a 5060 Ti, but at the time of testing the drivers on NVIDIA’s website weren’t playing nicely with Ubuntu 25.04, hence the 4060 Ti. 

DLSS took the biggest hit from Proton. On Windows, DLSS generated a 41.1% uplift in performance while the uplift was only about 32% on Ubuntu through Proton. XeSS also saw a decrease in performance from 37.1% on Windows to 32.7% on Ubuntu. FSR was the most surprising where, unlike DLSS and XeSS, FSR’s performance improved through Proton. FSR caused an 18.3% uplift on Windows, but it led to a 22.2% uplift on Ubuntu. Once again AMD came out on top for supporting their hardware on Linux.

1 of 2

Distro War 

I can already hear you typing “Um akshually, if you tried <insert your favourite obscure distro here> you would have measured even BETTER performance and Linux clearly would have smoked Windows in your tests.”, so I decided to test this claim myself. Using the 7700 XT I tested Black Myth: Wukong, Cyberpunk 2077, and Shadow of the Tomb Raider at 1080p across Bazzite, Manjaro, Ubuntu 25.04, and Ubuntu 24.04. In order to control as many variables as possible, I selected images that shipped with the Gnome desktop environment for each of these distros (I like how Gnome looks, please don't hate me). 

To say the results of this test were underwhelming would be an understatement. I measured near identical performance across all three of the tested games. The only distro that performed slightly worse than the others was Ubuntu 24.04, but even its results were well within what I would consider run to run variance - That 1 fps won't make or break your gaming experience. My takeaway was that as long as you select a recent distro, which supports your PC’s hardware, your out of box gaming experience will not be impacted. When choosing a distro, you should pick the one that suits your specific needs whether that be Linux Mint for the easy out of box experience, or Arch for the freedom to customize your system to your heart’s content. Regardless of your choice, you will be able to install custom tweaks to squeeze every ounce of performance out of your system, but exploring this topic in-depth would warrant its own article ;)

Conclusion

Gaming on Linux was quite the adventure, from weird graphical issues on the B580 in Ubuntu 24.04, to drivers that wouldn't install for the 5060 Ti on Ubuntu 25.04. My main takeaway from my testing was that if you want to game on Linux, trust the forums, save yourself the headaches, and select an AMD GPU if possible. The out of box experience on AMD is, in many ways, better than Windows, and you’ll never have to worry about a GPU driver installer again thanks to the amazing open source driver support. Furthermore, your performance in games will be comparable to that of Windows and you might even see some performance improvements in certain titles. Don’t be discouraged by the forums claiming X or Y distro is better for gaming, and that you absolutely shouldn’t use Z for gaming on Linux. Based on our tests, as long as you have a distro with an up to date kernel and the latest Mesa drivers your out of box experience won’t vary much at all.

If you have an AMD GPU, mostly use your PC for gaming, don’t need any proprietary DRM protected software like the Adobe suite, and are looking for an alternative to Windows 11, then it’s worth a shot. The great thing about Linux is that most distros will let you boot into the OS directly from a USB drive. This means that you can be test driving a new operating system on your PC within minutes. So why not give it a try? Going back to our original question: Is 2025 the year of the Linux desktop? As per the previous year, and the year before that; it’s up to you. The answer depends on your setup and your needs. If Linux can meet your requirements, then yes it is your “Year of the Linux Desktop”, but you’ll only know if you try.

Once again, we'd love to hear any feedback about our results and methodology. Feel free to visit our forum and let us know what you thought of this article. If you have ideas for Linux related things Labs could explore next, we'd love to hear them!