AMD Ryzen 7 7840U Windows 11 vs. Linux CPU Performance

[schmidtbag]

Linux fans who visit the general computer forums and the Windows forums always write that Linux is several times faster than Windows, and that Windows is so slow that it is unworkable.

The linux foundation pays them for this?? Is that what the donations are for?

Has open source religion brainwashed them?

It's more about day to day use and responsiveness. When you're just crunching numbers, the OS hardly makes a difference; the scheduler matters more. Where Linux tends to greatly outpace Windows is things like installations/updates, cold booting, and loading basic functions (like the Start menu or Task Manager). Windows carries a lot of bloat, which means it takes longer for it to swap things in/out of memory, which also has to happen more often since memory fills up quicker.

[edwaleni]

It looks like the latest kernel provides best Perf/Watt. That's awesome to see and aligns with work AMD has been doing on each new kernel. I think the only thing I've struggled with is how to limit my 6850u clockspeeds and turning off the core boosting stuff when I'm on battery and using p-state Active EPP. Some of the documentation I've read on how to do that seems to only work on Intel.

Have you looked at any of this yet?

https://docs.kernel.org/admin-guide/pm/amd-pstate.html

[skeevy420]

I can assure you he doesn't. He has stated that he sets all benchmarks up so that they are as automated as possible, remember all the Clear Linux tests where CL was tested with the performance governor and AVX512 compile time optimizations and distros would be tested with the power save governor and Windows with the balanced power settings.

Yeah, that kind of testing is what makes Phoronix actually Phoronix. It isn't Michael's fault that some operating systems come with settings that some people may or may not use.

Once you normalize all the systems so that they hit the same clock speeds, the performance differences largely go away.

Pentium 4 vs Zen 4 at 3.8Ghz

[citral]

It looks like the latest kernel provides best Perf/Watt. That's awesome to see and aligns with work AMD has been doing on each new kernel. I think the only thing I've struggled with is how to limit my 6850u clockspeeds and turning off the core boosting stuff when I'm on battery and using p-state Active EPP. Some of the documentation I've read on how to do that seems to only work on Intel.

If you grab the latest beta tlp 1.6.0 it allows to work with amd_pstate=active and you can set "power" for battery which greatly increases battery life, went from about 4 to about 6H on my 5600U.

[citral]

Michael has never mentioned whether he prepares Windows for benchmarking properly.

And that's a lot of work that needs to be done otherwise you could be getting quite low results.

https://www.techpowerup.com/forums/t...chmark.287480/

And? Nobody's going to do that except anal people who want to demonstrate something, and he doesn't "prepare" Ubuntu either. So it's a fair, out of the box A/B. What else now, should he make absolutely sure no forced update is running while doing windows benchmarks? Disable telemetry? At 3am at night when the system is done with all its background obscure stuff?

[citral]

Linux fans who visit the general computer forums and the Windows forums always write that Linux is several times faster than Windows, and that Windows is so slow that it is unworkable.

The linux foundation pays them for this?? Is that what the donations are for?

Has open source religion brainwashed them?

I disagree, when you're in since a long time you know that linux is primarily a fantastic system for admins, you can deploy whole infras with ansible and not worry much if something goes wrong with a server, can redeploy within minutes, and that it's the opposite with windows. Perf is not the primary concern there, even if it's way better.

When it comes to desktop, since have come a loooong way since year 2000 but it's far from perfect due to fragmentation, but overall I'd say there's less downtime due to various forced updates and other shenanigans, so it's more productive. When it comes to gaming perfs, we're seeing it catching up over time and are now within -10% for God of War to +10% for RDR2.

I don't know where it stands when it comes to stuff like video editing though, but codecs tend to go open source over time so things can't go worse than they used to, I imagine.

[Mitch]

If you grab the latest beta tlp 1.6.0 it allows to work with amd_pstate=active and you can set "power" for battery which greatly increases battery life, went from about 4 to about 6H on my 5600U.

That's pretty cool. I might try TLP. I was able to set my policy to power and my governor to powersave and p-state to active. The battery life is notably better and it runs much quieter and cooler. I'll see if either TLP beta or one of the other links someone posted can let me limit power. These CPUs are fast enough that I'd rather cap the clockspeed a little. IIRC, perf/Watt is better when you're not hitting the highest speed. I think the sweetspot was like 1.6 or something like that. For now, I want to try 2.4 since it's below my processor's base speed (2.7 base, 4.7 turbo).

[avis]

And? Nobody's going to do that except anal people who want to demonstrate something, and he doesn't "prepare" Ubuntu either. So it's a fair, out of the box A/B. What else now, should he make absolutely sure no forced update is running while doing windows benchmarks? Disable telemetry? At 3am at night when the system is done with all its background obscure stuff?

When you do benchmarking you want your system behavior to be predictable and tests to be repeatable.

If you don't prepare Windows for that, it may run certain background jobs unexpectedly (not limited to Windows Updates) and screw up your testing completely.

Linux doesn't need that because an average Linux desktop install almost has zero background jobs aside from system logging.

I wanted to end this message with some egregiously caustic remarks about someone's extremely intelligent comments but decided not to.

[citral]

I wanted to end this message with some egregiously caustic remarks about someone's extremely intelligent comments but decided not to.

Yeah, that's clever of you because you'd have looked like an idiot. It's not Michael's problem if windows doesn't perform its best stock. And in that sense, while I've been crictical of his latest risc-v benchmark, in a way it makes sense too, to benchmark what the manufacturer offers and not what a gazillion different tweaks could have brought. And can't be reproduced consistently.

[citral]

That's pretty cool. I might try TLP. I was able to set my policy to power and my governor to powersave and p-state to active. The battery life is notably better and it runs much quieter and cooler. I'll see if either TLP beta or one of the other links someone posted can let me limit power. These CPUs are fast enough that I'd rather cap the clockspeed a little. IIRC, perf/Watt is better when you're not hitting the highest speed. I think the sweetspot was like 1.6 or something like that. For now, I want to try 2.4 since it's below my processor's base speed (2.7 base, 4.7 turbo).

Obviously, past a certain threshold, it's diminishing returns territory indeed. Boosting is way past that, not something desirable on battery. At least a 15W TDP ryzen laptop needs at most a 45W PSU while boosting + max brightness + max GPU etc, the latest "6W" intel N200 needs a... 65W PSU. But they have better efficiency at idle to be fair.