I benchmarked my gaming PC on Windows vs Linux, and the difference was bigger than I expected

I’ve got a dual-booting PC with both Windows 11 and EndeavourOS Linux on it, which puts me in the best position to compare benchmarks across operating systems. It’s the same machine, after all, with the same CPU, GPU, RAM, SSD, etc. The hardware variables disappear when benchmarking on Windows and Linux, and it’s as controlled a test as I can get.

I’ve been dual-booting this MSI Cyborg 15 for a while now, though I do spend as much time as I can on the Linux side of things. I thought I had a reasonable sense of how the two OSes compared in terms of raw performance, until I ran the actual numbers. A couple surprised me.

The setup

Same machine, two operating systems

My MSI Cyborg 15 is a mid-range gaming laptop with a 13th Gen Intel Core i7, Nvidia GeForce RTX 4050 Laptop GPU with 6 GB VRAM, 16GB DDR5 RAM, and a 1080p 144Hz display. Windows 11 lives on one partition and EndeavourOS with KDE Plasma sits on the other. Both run the same Nvidia driver version 610.43.I ran both tests with the system on balanced power mode and nothing open in the background beyond what the OS has running by default.

I used Geekbench 6 for CPU and GPU scores, and Unigine Superposition for rendered graphics performance. They’re both cross-platform and have free versions, making them ideal for my situation. I was able to run the same workloads, using the same scoring methodology, on both OSes.

CPU performance

Closer than you’d think, with one exception

The results for the CPU were fairly identical with only one notable difference. In single-core performance, Windows came out ahead with a gap of about 8.5% (2,353 vs 2,170). Single core is what decides how reactive everyday tasks feel, and Windows has a consistent edge for things like file compression, browser rendering, and image processing.

Multi-core, though, tied things up. Linux scored 11,273 to Windows’ 11,188, a difference of less than 1% in Linux’s favor. The Linux kernel’s scheduler handles parallel workloads well, which shows up here on a 10-core chip like the i7-13620H inside my laptop.

So, basically, CPU performance is a wash, with Windows holding a small edge for single-threaded tasks.

GPU compute

Linux pulls ahead on paper

The GPU compute results were the most surprising, though, but also the most complicated to interpret. Running Geekbench 6’s OpenCL benchmark on my RTX 4050, Linux scored 78,007 to Windows’ 73,090, which shows a lead of about 6.7% for Linux. That means that my Linux OS did better at tasks like image processing, particle physics simulation, and feature matching. The Nvidia driver on Linux actually outperformed its Windows counterpart in my testing.

This isn’t unprecedented. Nvidia’s Linux driver has matured significantly, and for compute workloads that use CUDA or OpenCL directly, it can be competitive with or faster than Windows, and other dual-boot users running Geekbench on Nvidia hardware have reported the same pattern.

Graphics rendering

Where Nvidia on Linux still struggles

I tested rendered graphics with Unigine Superposition, a real-time 3D benchmark, that ran at an average of 87.65 FPS on Windows, but only 71.3 FPS on Linux. That shows a 23% gap in Windows’ favor (and it took some troubleshooting to even get there).

My first Linux run came up with 20.3 FPS, which didn’t seem right. A quick check with nvidia-smi showed the GPU sitting at 0% utilization even when I was running the test. Turns out, Superposition defaulted to the integrated GPU instead of my RTC 4050. On a hybrid GPU laptop running Linux like mine, you sometimes have to explicitly tell the system which GPU to use. Running the benchmark with two environment variables from Nvidia’s PRIME render offload documentation prefixed to the command fixed it, and the numbers jumped to a more similar result. If you’re running into the same issue, try running:

__NV_PRIME_RENDER_OFFLOAD=1 __GLX_VENDOR_LIBRARY_NAME=nvidiaEven with the discrete GPU accurately being tested, Linux trailed Windows by almost a quarter in rendered FPS. I used the same 1080p Medium preset, the same driver version, and the same hardware, which means that the broader community is right about poorer Nvidia hardware performance under Linux. Anticheat is still an issue, too, but for single-player gaming, a 23% FPS hit is the bigger concern here.

Idle RAM usage

A number that needs no explanation

I didn’t need a benchmarking tool downloaded for this check. I simply closed everything on each OS, let the system settle down for a few minutes, and checked my RAM usage with Task Manager on Windows, and the free -h command on Linux.

Windows 11 showed 8.5GB used while idling, while EndeavourOS with KDE Plasma was only using 3.4GB. That’s a 60% difference even without a single app open. Windows was sitting on more than twice the RAM just to run itself, which leaves less room for games, browsers, or anything else you might want to run on the machine.

What the numbers mean

Windows wins on single-core CPU performance and rendered graphics. Linux wins on multi-core CPU, GPU compute, and idle RAM — and wins the RAM comparison by a wide margin. The results aren’t a clean sweep for either side, which is probably the clearest answer anyone doing this test should expect.

The GPU rendering gap is the one that matters most for gaming, and 23% is significant enough. If you’re on an Nvidia laptop and gaming is your primary use case, Windows is probably still the better platform for raw frame rates. That said, the Linux gaming picture has been improving steadily, and game compatibility has reached the point where the remaining barriers are mostly business decisions rather than technical ones. For anything that doesn’t depend on the rendering pipeline, like productivity, development, and compute workloads, Linux holds its own and then some. For me, the RAM number alone explains a lot about why my Linux side feels a little nicer to use day to day, even though Windows wins the GPU benchmark.