WOW! Latest 5.7.12-1 kernel update dropped Ryzen 3950X temp a full 10° C

I always wait for Stable update notices here on Manjaro org. I have a handful of AUR programs (primarily Chrome) but for some reason Octopus notified me of a full update available. For some reason today was a full AUR update, not a common Stable Pacman update.

Anyway ... I rebooted into the new 5.7.12 kernel. I crunch both BOINC and F@H with 100% processors at a reduced ECO power level. Without a single UEFI BIOS change, suddenly my VCore temps are a full 10° and often 15° lower!

Surely it could be a coincidence, but it certainly is holding true on my Conky display for the past three hours since the reboot.

Does anyone have knowledge of the updates that might explain this? NOT COMPLAINING :smiley:

3 Likes

These are types of presents we get now that Linus has AMD :slight_smile:

4 Likes

maybe merge between linux 5.8 & 5.7 on this
https://www.phoronix.com/scan.php?page=news_item&px=Eight-Linux-5.8-Features

1 Like

"There is also other AMD energy/power work that's been happening recently as well ... "

I want to also confirm a 50KwH power reduction as measured by the UPS.
From 425+ down to 378

Hi!
Looks like the land of 5.8 will bring to the owners or AMD zen+/2 a lot of gifts

2 Likes

I don't get this. Idk what boinc and f@h are but if your processor suddenly dropped 10 degrees under %100 load, how is that ok?

Under 100 percent load a cpu should work at same temps regardless of the OS. Or do you think Manjaro plays with your clocks? That would actually mean this OS is dangerous, no OS should temper with base clock and voltage, and i don't think Manjaro does that.

To me this sounds like manjaro no longer is able to fully utilize your cpu.
You can always cut down max power from tlpui and it improves thermal performance greatly. But it's your choice not something imposed upon you.

I don't see any benchmark scores either so you're on to nothing actually.

I cannot claim that i know much about OS-CPU relations so if there is something wrong about this go easy on me.

1 Like

It's the kernel drivers. Not Manjaro-specific. I'd call them more efficient drivers.

1 Like

Efficiency has nothing to do with thermal performance though? Without benchmark scores you cannot determine efficiency, right?
I mean it should still work same temperature but should be able to pull off better scores on benchmarks. If it still has same scores with lower temps well that sounds great but there is no score here and there can be no conclusion.

I would like to see comparisons as well.

1 Like

Since i switched to AMD Ryzen 3400G a year ago i get very different H264 rendering times with Kdenlive.
10minute video rendered in ~15 minutes a year ago.
With last Kernels 5.0 to 5.6 it took ~30 minutes.
Since Kernel 5.7 the video finishes in 8 minutes.
Always experimental GPU/APU rendering enabled.
Maybe some CPU functions are enabled now?
:thinking:

2 Likes

Berkeley Open Infrastructure for Network Computing and Folding @ Home

Distributed computing where humongous mathematical problems are broken down to bite-size pieces where any consumer computer platform can sign up and take on a little piece in spare cycles when desktop demands are light.

I crunch the numbers in the background 24x7x364. With my settings prior to this kernel, I was heating up at 62-65°C VCore and afterward I am seeing 48-50°C

I do not understand at all your concern that Manjaro is somehow "playing with my clocks"

Every number cruncher is running at exactly the same production output I observed before, only now taking less power and generating less heat. The programs are running as well and as productive as before; only now burning less energy and power.

Not about benchmarks at all. Just a soldier in the trenches reporting how much lighter my burden suddenly became.

I am not really a lightweight in this arena:
https://stats.foldingathome.org/team/2233

What exactly is your objection to better productivity at a lower cost?

regards

4 Likes

F@Hx

Show me anyone else making 2 million PPD @ 50°C and I will shut up

Yes it does.
If the same process with the same resources and the same output is done in one example at a higher temp, then at another example with the same results but a lower temp .. I think we can safely call that 'more efficient'.
Whether you are talking about the software/firmware/driver/leverage efficiency, or the simple mechanics of 'when it works harder it creates a higher temp' to see a drop in temperature, but not production means it is working less hard for the same results .. meaning more efficient ..
I dont understand how that is even up for debate.

2 Likes

Have you read the rest of that post though?

I thought pointing out one fallacy was enough :wink:

1 Like

I have no objection. As i said so i didn't know what f@h was and i didn't know you had numbers there. Now that you have numbers and with the info pappl provided then this seems like good news indeed.

So what i understand here is that there is no playing with clock and voltage, it just doesn't require full load to run your f@h thing. But you say it's 100 percent so then it must be full load.

Hell, wth is this about? :confused:

I think pappl is right. They must have channeled the load to some other part, maybe this is all about a better floating point unit management or better use of cache. So while CPU seems to be working 100 percent in both cases the use of FPU must have changed.

I am using 100% of processors ... i.e. 16 cores and 32 threads. Not 100% power. As mentioned, I run on AMD ECO mode already in the UEFI BIOS in order to bring down the CPU temp. I am in a slow and steady marathon and do not want or need the Precision Boost nor running at full power. The only difference so far as I can tell is the new kernel.

1 Like

The plot thickens:

https://www.phoronix.com/scan.php?page=news_item&px=Google-RAPLs-AMD-Zen

Definitely a lot of activity on the Ryzen power front lately.

Forum kindly sponsored by