The YouTube channel “Maximum Fury” conducted a technical test of the new Cyberpunk add-on called “Phantom Liberty” on an older AMD hardware system, testing it separately on Linux and Windows 11. The Linux system, specifically the Fedora distribution called Nobara, performed significantly better, delivering 31% more frames compared to Windows 11.
The hardware used for testing included an Asrock B550 motherboard with an AMD Ryzen 5 5600 CPU and an AMD Radeon RX 5700 XT GPU from the first RDNA generation, along with 16 GB of DDR4 RAM. The CPU, RAM, and GPU were overclocked, and the system utilized undervolting to save energy costs.
When testing the game at 1080p resolution with high textures, the Linux system achieved an average of 63.72 frames per second (fps), while Windows 11 managed only 48.55 fps. This suggests that the game should run noticeably smoother on the Linux system.
This is a rare and extreme case, which is probably caused by some sort of fluke in the testing method or due to a bug in the game that Linux is handling better. Usually gaming on Linux is like ~5-10% slower for GPU-bound games.
This is likely going to change as software support for gaming on Linux improves.
If you consider real high performance computing, with well optimized libraries that can properly use the hardware (including GPUs), 50 % difference between windows and Linux is not really surprising. This is the reason 100% of real high performance computing is done on Linux. It is a better OS for raw performances than windows. For some tasks we are easily talking over twice the performances. It is not always the case, but not surprising at all.
The differences clearly depend on the actual low level implementation of the code. But in general the current situation in gaming, with windows that competes with Linux on raw performances, is only due to lack of software support for gaming on Linux. As this is changing over time, we’ll see games performances greatly improve in Linux. Hopefully until the physiological surpass of windows performances.
Currently most of gaming support on Linux is done via some kind of translation layer, that has itself an overhead. It means that the real linux performance would be even better than in all these benchmarks, if it was really possible to compare 1:1 Windows and Linux with native, well optimized code.
This is probably more common than you’d think, at least in my anecdotal experience. Converting directx commands to vulkan commands, especially for AMD GPUs, can result in better and more consistent performance on Linux.
Do you have any numbers or examples of games? I know that it’s generally the case that DX9 games often have greater performance through DXVK, but DX11 and DX12 should usually be a little bit slower. Also, CPU-bound games are often faster on Linux in my experience, but it’s rare for games to be CPU-bound (MMOs etc).
Additionally, OpenGL and Vulkan should be faster on Linux (Native or WINE+OpenGL/Vulkan), but I don’t have as much experience with them.
Edit: I found this video which has a few standout games where Linux pulls ahead even on DX11/DX12. Hopefully that’s a sign of future trends.
There was a tweet before the recent Cyberpunk update that essentially said “expect very high CPU utilization as we now use the whole CPU” which I thought just meant they dropped the ball somewhere.
I haven’t done extensive testing on this as I’m just some dude. It’s been a long time since I’ve had windows running on anything, but the three that I remember are:
Or faster. Depends heavily on the game. Some things wine + dxvk does better.
Sometimes there are also unimplemented/broken features on Linux which people don’t notice and save frames. Legit performance improvements over Windows do happen (especially on memory and cpu-limited systems) but I’d be skeptical of any particularly huge ones.
Yeah.
I’m personally lucky that my fav titles are CPU hogs, like ARMA 3 and X4: Foundations. Both run better under Linux.
Cyberpunk runs great too, I’m sure once we eventually get the updated drivers for NVIDIA we’ll get Ray Recon too.
Can relate 🤓
Only thing I’m missing is “real” head tracking. There is simply none in the Linux version and while I can map a virtual joystick driven by OpenTrack to each camera corner it’s just not the same. Sadly this is not exposed via LUA or I’d have wired up a UDP connection by now. So this feature sadly works only via Proton. Still sticking with the native Linux version though. It’s faster.
It’s not rare for games to be a few % faster, as long as they’re using features that are well supported in Linux. If the bottleneck is something that needs heavier emulation because the native implementation isn’t available or good enough then yeah you’ll see slowdowns.
I kind of expect a patch for Windows that addresses the reason it is slower there now that they know there is a difference.
On Nobara OS, I haven’t noticed any performance dip coming from windows.
Linux Experiment on youtube found it performs ~5% better overall in games than Fedora, so that’s probably why.