Google Posts Patches Allowing AMD Zen/Zen2 CPUs To Expose Power Usage On Linux Via RAPL

[_Alex_]

Maybe check a few more distros with live CDs/flash to see if they do something differently regarding idle. And if you are into custom kernels, maybe eliminate cpufreq and cpuidle susbystems altogether, allowing the cpu to undervolt/underclock on its own based on the loads it detects, rather than having the OS algorithms control it and do something wrong - if they are doing something wrong (?).

+15w compared to windows seem a lot. Personally, on my 2200g (ok not a zen2 obviously), windows are far worse in terms of power consumption at idle.

Btw, regarding power usage on linux, I just discovered by accident that the turbostat program exposes core watts and package watts. However, I think the numbers are probably off by quite a bit. For example on my 2200g I'm idling at ~17w, hitting 67w in prime (on the plug meter)... but turbostat shows 10.18 CorWatts and 18.53 PkgWatts... it may need an adjustment in the variables.

[vb_linux]

[QUOTE=Volta;n1179748]

It's AMD. Maybe AMD should fix intel's frequency manager? Furthermore, intel better fix their CPUs which are spyware.

This guy trolls at another level. His excuse may have been replying to another troll. Though, I have always seen him jump to defend AMD and then troll super hard.

How are AMD CPUs not spyware and Intel's are? Any logic behind this.

Currently, intel performance is at pentium II level.

Sure.

[vb_linux]

This is just embarrassing. And now they can't support agesa code on pre-500 series AM4 MB's. Their software dev really went sideways when they shut down their German software team.

AMD has the hardware and net profit margins now. They need to unfk this 3rd world software dev they have going on. That is now AMD's achilles heel, next to their Chinese production exposure.

Well, but isn't this the advantage of open source strategy. Users do not have to depend on AMD to get the best out of the hardware.

[fuzz]

IMHO, the company using AMD GPUs and CPUs for their streaming game service having a vested interest in ensuring said AMD hardware works properly isn't as interesting as it is obvious.

You read too much into my statement.

[Guest]

I hope, this will make the power consumption lower with forthcoming patches. I have a wall meter here and the idle power consumption of my 3700X is obscenely higher than in Windows. Newest arch kernels, of course. I get ~80 W idle here AT BEST (RX 5700 XT + ROG Strix X570-E Gaming) – normally it’s around 100 W. Whereas under Windows, it goes as low as 65 W. That’s downright embarrassing for a company selling server chips.

Maybe it's the GPU? I noticed my R9 390 idle power consumption is much higher on Linux than in Windows. Even under load, the power consumption is much higher on Linux than in Windows. I know it's a different one, but maybe it's the same problem?

[Guest]

The whole point of supporting the free software movement is that AMD can focus on building the hardware, and the community can help contribute to the software. This is win-win. Why wouldn't AMD be glad that Google is helping them?

Lol, no. AMD is a commercial company selling a product. I don't want to spend hundreds of dollars on a product, to be given the privilege of writing code for them for free. I'm not going to do their work and make their product work for free. They'd better be paying me USD 120000+ for that privilege.

[atmartens]

Lol, no. AMD is a commercial company selling a product. I don't want to spend hundreds of dollars on a product, to be given the privilege of writing code for them for free. I'm not going to do their work and make their product work for free. They'd better be paying me USD 120000+ for that privilege.

Right, but apparently Google engineer Stephane Eranianat is.

[Guest]

Right, but apparently Google engineer Stephane Eranianat is.

I'm talking about when consumers buy AMD hardware and get bad software support. Google only cares about Linux on AMD as far as servers, Stadia and ChromeOS. Not everything they do will help end users.

[pal666]

Well, but isn't this the advantage of open source strategy. Users do not have to depend on AMD to get the best out of the hardware.

amd provided best on windows. linux provided it on linux

[boxie]

It is already set to schedutil, conservative doesn’t change a thing.




Already did Powertop. Besides from some annoying problems this causes, no significant difference.
And I’ve had this excess in power consumption between Linux and Windows with a Radeon RX 580 too, it’s not the AMDGPU driver for the RX 5700 XT causing this (tested with light GPU load on Windows too).
But thank you both guys for your help, seriously.

I actually think, it’s AMD not getting their powerstates fully upstreamed.

No good sir, it is I that must thank you - I have now got radeon profile up and running and it is wonderful