I believe it, Windows bloat these days is so bad. I keep telling my friends Tarkov runs better on Linux if they'd just let me play the goddamn multiplayer I'd be golden
post
Take aways:
- Sample set is of 5 games
- Lenovo drivers are much slower than Asus
- There are 2 games where windows is neck to neck or better, 3 where SteamOS is far ahead
Some doubts:
- Did the author run the benchmarks few times to rule out shader compilation. 99%ile would be helpful.
- I wonder if it makes sense to test DirectX10, 11 and 12 games separately to better understand where Proton has an edge.
- I wonder what all settings can be tweaked in Windows to find potential fixes (core isolation, cpu boost, power profiles).
Point is Microsoft and OEMs need to do better, however not every game or subscription services work on Linux, so in the interim time users should know what they can do to close the gap better.
Did the author run the benchmarks few times to rule out shader compilation.
Why should the author rule it out? Honest question. If shader compilation leads so worse real world experience for gamers on Windows than SteamOS, it is a valid point to include.
Because I'm more curious about why things are the way they are just like the author, and would like to understand this with more data points, only making the comparison more helpful. I'm not saying author "should" consider impact of shader compilation, but I'm saying had they done, we'd understand the difference better.
They added asus vs Lenovo drivers data points, which alone tells us that driver optimization is responsible to a great extent. All I'm saying here is more data is more helpful.
Maybe even after taking care of that, the difference is huge, which will tell us its not enough to have precompilation of shaders. Maybe it does reduce the gap, telling us that potentially dx11 games might tend to do similarly.
Saying "RTX 5060 is better than 9060 XT" with 5 games tested is one level of comparison, but if they are grouped into RT and non RT games, games with 8gb and 16gb VRAM requirements, games with and without nVidia partnership, isn't that just more detailed and an even better comparison point?
This is really not surprising to anyone who has used modern windows and Linux recently. Windows is so incredibly bloated, whereas Linux is a true real-time OS basically out of the box.
Unless you use an RT kernel, Linux is not a realtime OS and certainly not a true one.
Because, you know, terms have a meaning.
Right but switching to an RT kernel is trivial for basically any mainstream distro. You can do it from the package manager.
True, I just wanted to clarify that by default Linux doesn't run on an RT kernel.
And tbh, an RT kernel is really not desirable for most applications, which is why it's not default. All these RT guarantees cost a lot of performance, and in most cases a guaranteed latency is not worth losing performance over.
In fact, using an RT kernel would be just the opposite of what you'd want on a gaming system.
While the bloat exists, even debloated windows wouldn't match proton because that's not the only reason. Despite bloat there are two games in this test the actually do similar or better than SteamOS. This means there's a confounding reason for the difference, not the bloat.
I recently switched from windows (with a debloat scrpit ran on it) to linux mint and I was shocked at how much faster it booted. When I turn my pc on I usually get up and do something else for a bit (not because windows is THAT slow but because I could spend the minute it takes to turn on to make lunch or something) and linux booted before I was out of my chair.
Did the author run the benchmarks few times to rule out shader compilation
Really grasping at straws there, eh? I'm no big fan of Ars but I hope we can assume they're not quite that incompetent.
Methodology is important to a robust result. It's weird that you take issue with their considerations there.
It's not a slight, as I said it's a doubt, not criticism. I'm not saying "did the author EVEN ..."
Your other doubts and concerns seem slightly biased, e.g. wondering what settings could be tweaked on only one of the systems being tested and then reminding us all that there do still exist some things that won't run on SteamOS. It's only that one that is outright ridiculous.
Biased to what? Point of comparison is to figure out why things are the way they are and use that information to get the best of both worlds? It's not very helpful if the conclusion stops at "x is better than y".
Going deeper into "why" Proton is doing better in 3/5 games but not in 2/5 will only help users of both operating systems to make better informed decisions and get everyone closer to root cause other than "bloated windows" or "just use linux", potentially even leading to improvements to both sides.
I found the same thing on CachyOS (another Arch fork). The increase for me was staggering. Lies of P went from an unstable 144fps on windows 11 with an overclock (OC) on my GPU to 200fps in Cachy. Settings were all maxed out at 1440p. I noticed a similar jump from other games. Modded and vanilla NBA 2K25 went a stuttery mess at 172fps (frequent dips down to 72fps) to a steady 180fps with NO dips (that’s my monitor’s limit). I like to test things on The First Descendent, and it went from an unstable 79fps with maxed settings to 119fps. And while I don’t have numbers for it, The Witcher 3 Next Gen (vanilla and heavily modded) run a lot smoother. But after ten years, that game has been optimized out the ass.
I did notice, however, that the increase in performance diminished greatly as I turned down settings. On Windows 11, I would notice a way “higher” increase in frames. For Example, I could tweak settings in the First Descendent like Global illumination and increase frames in Windows 11 to 109fps, but still unstable. In Cachy, if I did these things, I didn’t really notice a meaningful impact.
RT also performs slightly worse on Linux. But I figure anyone using Linux might be the same type of person to not care about RT.
My hypothesis is that without the CPU resources being eaten up by things like Windows Defender, the CPU is able to process more data quicker, reducing GPU wait time. I don’t have data on that, I would need something as in depth as presentmon from Intel for testing. Arch has forks of that, but nothing nearly as in depth, and PresentMon has declined any Linux support in the foreseeable future.
I should mention, the OVERALL jump is ~40% going to CachyOS. And we know that the jump from Windows 10 to 11 saw a ~27% hit due to the new Windows Defender.
My system is 64GB of SK Hynix DDR5, 9070xt (on my Windows Partition it’s OC’d, but on CachyOS I leave it stock), and a 9800x3D that has been manually OC’d in the bios and a 240mm AIO. I leave the panels off my O11 D Mini. The motherboard is a Gigabyte X870 Aorus Elite (2x8 pins for the CPU delivery).
For all the FPS data, I pulled it from Steam on Cachy which uses presented frames instead of actual frames. Basically, the frames the GPU is presenting to the monitor, not necessarily what your eyes are seeing.
On my Ally, I also noticed a difference swapping to SteamOS. Something to keep in mind with anyone planning to do that, you can allocate up to 6GB of RAM to the iGPU before Arch/SteamOS gets affected. I just don’t see anyone telling you you can do this.
Edit one day later- I played Enshrouded on CachyOS. I will report that my 9070xt underperforms at max settings. Unstable 80fps with dips down to 50fps, but the Frame Time Pacing makes it feel worse. It stutters like it’s running at 50. Turning down settings again only increased frames by 5fps, which is not marginal at these rates, but did not help with the stuttering issues. I think it’s rendering things similar to Minecraft. The comparison I have is my 7800xt, which at max settings a year ago was able to run in Windows 11 at 70fps, but equally unstable. Therefore, I’d hypothesize that if I ran present on I’d just see high GPU wait times.
RT = Rollercoaster Tycoon?
"Hello everyone, and welcome to another video"
- Marcel Vos
You could use Nsight, it has a Linux version and is very in depth (shows every draw call, also has one that shows very detailed CPU tasks)
Of course harder to use than presentmon
Games run faster on SteamOS with proton than Windows 11, Ars testing finds
FTFY. I hate all these articles that downplay the heavy lifting proton (and all the tools that make it up) are doing. But "Proton makes games run better" doesn't get the same attention.
Proton is amazing, but it's entirely overhead translating library/system calls to Linux. It's accurate to say they run better on SteamOS, not to say Proton is making it run better.
Now maybe Proton makes them run better than a janky but native Linux port, but that's a separate statement about games being better optimized on Windows.
Hogwarts legacy, which is a exe, runs on proton but not on windows 10. I'd say proton runs better than windows.
They're not only being better optimized on Windows which results that running them through Proton is better. In a lot of cases Windows versions actually run, while native Linux don't, because there's no single stable API (ABI? Idk) on Linux and games break when you update your system.
Proton is amazing, but it’s entirely overhead translating library/system calls to Linux.
That is not at all true.
but that’s a separate statement about games being better optimized on Windows.
Is that though? You can't say X is better than Y when you're changing multiple variables. If windows had a proton equivalent and both games ran through it then yes that would be a direct comparison. But you can't say X + Y is better than Z (by itself)
DXVK is a part of proton that also is available on windows. DXVK alone can get you double digit performance improvements on games. And that's not getting into all the one off tweaks users can do to proton to optimize the game. Enabling pre compiled shaders gave a huge performance boost for Elden Ring.
The compatibility layer is overhead, but the key difference for many games is that DXVK swaps directX for Vulkan, and Vulkan often gets better performance.
The performance gains of using steamOS are twofold, there's less OS load (this is particularly noticeable in low performance games, windows will consume much more battery on a game like Dead Cells than SteamOS will), and there's also a vulkan performance increase for some games. My understanding is if you see a big performance increase in a demanding game, that's usually thanks to vulkan.
With your theory, you could run a DirectX to Vulkan translation layer on Windows and also get increased performance. Which may be true, but once again points the finger at bad drivers.
Yes, from what I've been told that actually does improve performance in many games.
There is overhead but Vulkan allows you to batch draw calls in a far more efficient manner. It can also generally use multi threading to feed a GPU even if the game isn't coded with that in mind. Basically Vulkan offers so many improvements to efficiency and parallelization that the overhead is a drop in the bucket compared to the overall speedup in draw call optimization alone.
In the same way that talking to a presidential translator is faster for a diplomat than talking to Trump. The translation layer can communicate more concisely and effectively.
But you can't say X + Y is better than Z (by itself)
I mean, yeah, you absolutely can. Especially when X + Y and Z are both common configurations, and using X or Y by themselves is uncommon or a known bad setup.
Sure, you can't be certain which of X or Y is making the differences in the comparison, but the comparison can absolutely be made.
I’m not sure it’s a Wine/Proton thing, it’s quite likely to be suboptimal at some things because it’s reverse engineered (not to diminish technical marvel that it is and decades of effort). Regular desktop Windows has just way too much overhead coming from everywhere.
As a side note, back in the day when Nvidia released drivers for FreeBSD using Linux binary compatibility layer was even faster than Linux for gaming.
Would love to see tests like this attempting to use DXVK etc (as part of their testing on Windows) to better isolate more factors
Yes but Microsoft Teams runs like dogshit on my Linux laptop. Checkmate atheists.
😭😭😭 Sadly, Microsoft Teams runs like dogshit everywhere
But have you tried Outlook (NEW) and Teams (NEW)?? Microsoft made changes to deeply integrate copilot into them, while making the UI unintelligible and broken as well. It's a much more authentic Windows experience
I was using Teams on Firefox and I've been thinking that FF is the problem that Teams's working horribly, but man, it works horrible on chromium as well
We'll have to see if that's the same with the Xbox Ally.
I'll be laughing if its still outperformed
The XBox Ally could still be worthwhile if it ran XBox One and backwards compatible XBox 360 and XBox games. I don't know why Microsoft didn't do that.
It is more impressive when you realize that those games were meant for Windows and require a translation layer (Wine and DXVK).
Preach. Studios that make games with anti cheats and what not should reconsider how they handle Linux as they'll only get even more players, who'll probably be even more loyal due to their Linux compatibility. I know cheating is a big issue in online games, but adding invasive kernel level code to detect that is just adding system level vulnerabilities just to prevent cheaters from cheating seems like an overkill. It's not like cheating mouse and keyboards don't exist and cheaters have evaporated entirely due to anti cheat.
top 99 comments