Wednesday, August 19th 2026

CachyOS and AlmaLinux Beat out Windows 11 in Workstation Benchmarks

With the advent of the Steam Deck and the bump in Linux adoption that we've seen since then, a number of Linux distributions have appeared that make tweaks to the Linux kernel in the hopes of gaining performance. CachyOS is one such Linux distribution, and it appears that those kernel performance tweaks are paying off, according to some recent testing by Phoronix.

In a comparison between CachyOS, AlmaLinux 10.2, Ubuntu 26.04, Fedora Workstation 44, and Windows 11, CachyOS consistently performed at or near the top of the list, with CachyOS in performance mode claiming first place in the geometric mean of the results of a workstation benchmark. Windows 11 Pro was right at the bottom of the chart, coming in 14.9% slower than CachyOS in performance mode.
It's worth reading the full write-up at Phoronix for the in-depth results, especially if you're using benchmarks to determine which operating system to use with specific software, because there are enough cases in which Windows scores a decent victory that an application-specific approach is warranted.

The tests comprised of a mix of CPU and GPU benchmarks on a $46,000+ HP Z4 G6i workstation PC with a 48-core Intel Xeon 678X Granite Rapids WS CPU, 128 GB of DDR5-6400 memory, and an NVIDIA RTX Pro 6000 Blackwell workstation GPU. First up was 3D rendering and modelling, which tested Blender and IndigoBench 4.4 in both CPU and GPU rendering modes. Here, almost surprisingly, given NVIDIA's reputation in the Linux community, CachyOS and AlmaLinux were both faster than Windows 11 in both CUDA and CPU compute. The biggest differences in the render workloads came from CPU-only rendering, where Windows 11 was up to 12% slower than CachyOS. Ubuntu and Fedora were both measurably slower than CachyOS, but only by about 3-4%.
Tides turned with Chaos Group's V-Ray 6.0 rendering engine, where RTX testing resulted in 6% more rays on Windows, while CUDA rendering resulted in a 2% advantage for Windows. In CPU rendering, Linux was once again ahead, with CachyOS leading the pack at 18% faster than Windows but only marginally faster than the other Linux distributions tested. Neatbench video benchmarking, GravityMark 1.87, Unigine Superposition 1.0, and OctaneBench also saw CachyOS pulling ahead, although in some tests, gains were minimal, and in Octane, only CachyOS in Performance mode and AlmaLinux managed to pull ahead of Windows. In FluidX3D fluid dynamics simulations, performance was largely identical, but Windows 11 did manage to technically edge out a lead, even if only by fractions of a percent.
OSPRay Studio and LuxCoreRender produced peculiar results, with Windows matching or edging out Linux in a number of tests but losing badly to all of the Linux distributions in others. 7-Zip compression benchmarks once again saw Linux beating Windows by a decent margin, although it was basically a wash between the different Linux distributions.
Source: Phoronix
Add your own comment

87 Comments on CachyOS and AlmaLinux Beat out Windows 11 in Workstation Benchmarks

#76
GeoffreyA
AshinaGravychiroWell yeah that's what I'm trying to say. Win32 is old so it will run fast in modern hardware. Or is there more nuance to this?
Even if you compare something platform neutral like Qt. It still runs noticeably slower than on Linux. So I won't say it's just a win32 issue imo.
Win32 is old, certainly, but importantly, low level. Microsoft's later .NET Framework was slower because applications executed on a sort of virtual machine, the CLR, similarly to how Java bytecode runs on the Java VM at runtime.

Being built on COM, WinRT is supposed to be a native approach: at least in the case of C++/WinRT. C# and other languages have their objects repackaged at runtime for interoperability with the ABI, entailing a cost.

I always thought that Qt runs sluggishly because of wasteful C++ usage, particularly dynamic allocation. I am not sure why it would be faster on Linux. Perhaps Qt's implementation on that platform is better.
Posted on Reply
#77
AshinaGravychiro
GeoffreyAWin32 is old, certainly, but importantly, low level. Microsoft's later .NET Framework was slower because applications executed on a sort of virtual machine, the CLR, similarly to how Java bytecode runs on the Java VM at runtime.

Being built on COM, WinRT is supposed to be a native approach: at least in the case of C++/WinRT. C# and other languages have their objects repackaged at runtime for interoperability with the ABI, entailing a cost.

I always thought that Qt runs sluggishly because of wasteful C++ usage, particularly dynamic allocation. I am not sure why it would be faster on Linux. Perhaps Qt's implementation on that platform is better.
Well I say that but still I think the Windows Terminal boots pretty quick for a WinUI application. Maybe the implementation also matters I guess.
Also Qt runs apps don't use C++ properly?
Posted on Reply
#78
lexluthermiester
AshinaGravychiroWhat would be a debloated / tweaked Windows? What do you personally do on your systems?
Win11, customized of course. But I also run ZorinOS on a spare desktop, Linux Mint on one of my laptops and AndroidX86 on another.

EDIT: Forgot to mention ZorinOS..
Posted on Reply
#79
MTS-C
LittleBroDepends how powerful the machine is. 4c/8t with 8GB RAM will be slowed down dramatically. 8c/16t with plenty of RAM not so much.
However, slow context menu and similar stuff will affect any PC.
Like I said, 23H2 so no alleged speed ups. There is no slowness in responsiveness to core UI objects compared to say my main rig which is on 26H2 insider preview.
When I started at this particular company my initial assigned asset was a Surface Laptop 4. It had the 1135G7 and 16GB of RAM. I quickly had to have a jump box for us spun up with enough RAM to RDP into many different systems at once because their base image bloat was hogging more than half of the available RAM causing RDP sessions to crash or the RDC manager software to crash when the OS would try swapping. This was while still using W10. I was part of the testing group to move to W11 21H2 summer of 2022. There was no perceivable slowdown back to back with everyday tasks.
My current asset should technically in theory be faster than my main rig as it's a Dell Pro Max that cost around 7000 (after fake shortage pricing adjustments) with a Core 2 Ultra 9 285H and 32GB 8400 RAM but it is definitely bogged down by their bloat. Dinking about in Windows is quick and responsive, loading a program or moving files is not.
Posted on Reply
#80
Tess
GeoffreyAThere's a big question mark on this test. Is VBS enabled? Which it usually is by default.
Yeah. There are half a million settings in both operating system and BIOS/UEFI which can make a substantial difference in performance - even those which may favor one or the other. Without knowing them it does leave a bit of question mark hovering over the results.
Posted on Reply
#81
GeoffreyA
AshinaGravychiroWell I say that but still I think the Windows Terminal boots pretty quick for a WinUI application. Maybe the implementation also matters I guess.
Windows Terminal is C++/WinRT, hosted on a Win32 window, so it should be on the faster side.
AshinaGravychiroAlso Qt runs apps don't use C++ properly?
I don't have enough experience in Qt to answer, but I remember when I learnt it, the idiom was to allocate controls dynamically, and while the code was elegant and clean, it was clearly heavy during execution.
Posted on Reply
#82
Grimling
Every Code, that is not HANDWRITTEN on a binary level, is SLOP!
Posted on Reply
#83
AshinaGravychiro
lexluthermiesterWin11, customized of course. But I also run ZorinOS on a spare desktop, Linux Mint on one of my laptops and AndroidX86 on another.

EDIT: Forgot to mention ZorinOS..
I meant as in , what tweaks do you apply to your win11 system
Posted on Reply
#84
lexluthermiester
AshinaGravychiroI meant as in , what tweaks do you apply to your win11 system
Quite a few. I start with either 11Pro or LTSC, depending on certain factors, I remove programs and feature I either don't need or don't want and customize from there.
Posted on Reply
#85
AshinaGravychiro
lexluthermiesterQuite a few. I start with either 11Pro or LTSC, depending on certain factors, I remove programs and feature I either don't need or don't want and customize from there.
I see, do you use Titus's winutils?
Posted on Reply
#86
Bloste
GrimlingEvery Code, that is not HANDWRITTEN on a binary level, is SLOP!
Dont't you mean "Perforated onto the punchcard"?
Posted on Reply
#87
lexluthermiester
AshinaGravychiroI see, do you use Titus's winutils?
Nope.
BlosteDont't you mean "Perforated onto the punchcard"?
Nope! Don't be daft..
GeoffreyAAgreed. However, I do think there are problem areas in the shell, springing from the use of XAML, WinUI/RT, managed code, etc. Other than that, the OS is fast.
I seem to have missed this post, however, that's a rabbit hole best avoided in this thread.
GeoffreyAThe NT kernel gets blamed for being bloated because of bloat in excess out-of-box apps.
Which is unfair as the Linux kernel has just as many parts to it some would consider "bloat". The reality is, in both cases, those parts of the kernel contain code needed for various reasons, even if uncommon, that preserve something most people don't place enough value on: Backward Compatibility. There is just as much of it needed in Linux as there is Windows. It's very important! Many fail to grasp this point and thus they complain needlessly about things they do not understand, either in full or in part.(and none of that was aimed at you, I'm just saying in general terms)
Posted on Reply
Add your own comment
Nov 13th, 2026 22:49 CST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts