AMD Ryzen 7 7800X3D: Windows 11 vs. Ubuntu 23.04 Linux Performance

[ll1025]

Hmm, 7%.

Linux fans claim that Windows is so slow that it is impossible to work on it.

They claim that on Windows all operations take several times longer and Windows takes all resources and for programs nothing is left.

And it turns out that in the right 28% of the programs Windows is even faster

They're also testing Windows with VBS enabled which (I believe) virtualizes the entire OS. I've never understood why phoronix does this because Linux has no equivalent security measure and it seems rather biased.

[ll1025]

It turns out Linux eats windows for breakfast even if one of the slowest Linux distributions was tested. 7%? In some benchmarks it's nearly 50%. It seems winboys are always out of touch with reality.

Only if you don't pay attention to the test setup. Turn off VBS / HVCI -- put windows on the same level of mitigations as Linux for an apples-to-apples-- and let's see how it shakes out.

[ll1025]

AMD announced that back during the reviewer briefing... But I only tested the 7800X3D on Windows anyways. I always do fresh installs.

Michael,
Can I ask why you use VBS for these tests? AFAIK it is not mitigating a specific issue the way IBRS is and I'm not aware of any equivalent feature in Linux. It's also not universally enabled, the way spectre etc mitigations are.

[Michael]

Michael,
Can I ask why you use VBS for these tests? AFAIK it is not mitigating a specific issue the way IBRS is and I'm not aware of any equivalent feature in Linux. It's also not universally enabled, the way spectre etc mitigations are.

Each OS was run at its defaults.

[Anux]

Michael,
Can I ask why you use VBS for these tests?

For the very same reason all distros get tested in their standard config. Michael has explained that over 100 times, it's what the typical user uses. No standard user is going to tune it's linux or windows to achieve maybe 3 to 5 % better performance, that is only for us nerds. Also there would be endless debates which tuning settings he should use.

BTW what would you expect from deactivating VBS? As far as I understand it has to do with driver loading and integrety checking (SecBoot, TPM, etc). Nothing you would see in a user context benchmark.

[Jabberwocky]

If I could burn all the strawmen in this thread I could power my house with it.

The 7800X3D is a complex chip to use to it's full potential. I would have assumed AMD would spend more time working on Windows scheduling (as seen in the past) and that it would pay off.

Paradigm Shifter I agree with different hardware and telemetry but even that impact varies between different hardware configurations. The 7800X3D can handle a ton of spyware and still complete the job in an instant (at the expense of initial-cost/heat/electricity consumption). The more complex a system is the slower it become too. That is my experience with VSCode. There's so much that can go wrong that prevents you from having an optimal experience. On my Linux laptop it's memory limitations. On my Windows desktop it's slow/unstable hardware acceleration of Blink (rendering engine of Chromium/VSCode). I still use VSCode daily because of the amount of community provided bells and whistles it does some jobs faster while being sluggish on a 24-Thread@(4.0 to 4.8Ghz) 32GB RAM tightly clocked system.

Smurphy that middle paste mouse button, I cannot agree more. That combined with the clipboard manager was a huge productivity boost! I find it funny that they mostly use macOS and Windows in design classes but limit people to a specific way of thought making it very difficult to change and explore designs. Linux has had compiz beryl superkaramba and newer things like https://github.com/letoram/safespaces which gives people so much more to design and innovate.

As for the boot time arguments it's all apple and oranges in most cases. Linux with non-standard stack can boot in less than 1 second. In general Windows boots really fast without making any changes to it. Some distros take really long to boot up and require you to enter your password multiple times during the boot process. If you want an apples to apples comparison then you need to use the same hardware with the same level of security, if one OS uses a TPM then the other OS should also use a TPM.

[mdedetrich]

Its well known that Windows default filesystem NTFS is quite a lot slower than the common Linux filesystems (although this is largely due to design/backwards compatibility reasons, trust me, Microsoft has tried as much as possible to make NTFS as fast as it can be). However as said in this thread, once files have been loaded into memory there really isn't a big performance difference. Both NT kernel and Linux kernel can be tuned to be as fast as eachother and you get to a point where you just split hairs.

Also note that you can use Windows with different filesystems, i.e. there is a BTRFS project for Windows i.e. https://github.com/maharmstone/btrfs

[ll1025]

For the very same reason all distros get tested in their standard config. Michael has explained that over 100 times, it's what the typical user uses. No standard user is going to tune it's linux or windows to achieve maybe 3 to 5 % better performance, that is only for us nerds. Also there would be endless debates which tuning settings he should use.

BTW what would you expect from deactivating VBS? As far as I understand it has to do with driver loading and integrety checking (SecBoot, TPM, etc). Nothing you would see in a user context benchmark.

Perf difference is frequently 5-10% difference but as much as 20% in some cases and on some hardware. HVCI / VBS can significantly impact performance because they virtualize the entire operating system, and segment off a separate high-trust kernel, along with a lot of memory translation overhead.

The problem is that it is not an apples-to-apples. You could find a distro that turns all of the recent spectre+ mitigations by default and it would mop the floor with windows-- but it is not representative of the OS's efficiency or design, but rather of the heavy impact of security in 2023. You can just browse through this thread and see all of the claims that this definitively proves that Linux is a better OS while ignoring how much hardening is absent. And to claim it's a default is to ignore that frequently VBS is turned off for various reasons.

By default Windows uses Secureboot + TPM + Bitlocker. Would it be reasonable to IO benchmark that "default" against a linux distro not using LUKS? Of course not, for apples to apples you want to test the kernel + drivers, not the encryption overhead and any implementation differences.

[flower]

Perf difference is frequently 5-10% difference but as much as 20% in some cases and on some hardware. HVCI / VBS can significantly impact performance because they virtualize the entire operating system, and segment off a separate high-trust kernel, along with a lot of memory translation overhead.

Windows Virtualization has an up-to 20% perf penalty? That really sounds too much (or they need to optimize it)
I've played for years in a windows vm running on a linux host and never saw a performance impact that big.

only thing that really had an impact was heavy io

[mdedetrich]

Windows Virtualization has an up-to 20% perf penalty? That really sounds too much (or they need to optimize it)
I've played for years in a windows vm running on a linux host and never saw a performance impact that big.

only thing that really had an impact was heavy io

Yeah that makes zero sense. Windows Virtualization is heavily used in Azure, an area where Microsoft cannot afford a 20% perf penalty since that is their biggest sources of revenue.