Linux 5.11 Released With Intel Integer Scaling, AMD Performance Boost, RTX 30 KMS

[kripteks]

After installing 5.11 on an AMD Ryzen 3800x my CPU core speeds--both max and variance-- have increased markedly.
Checking /proc/cpuinfo I rarely saw CPU speeds greater than 4.5 GHz before this kernel. With 5.11 I'm seeing max single CPU freqs exceeding 5.3 GHz. Most of the remaining CPUs stay ~ 2.2 with minimal demand (not idle, typing this message) . Running a VM puts the 4 allotted CPUs in 4.2- 5.4 range when I dictate into DNS in a Windows 7 virtual machine. something seems to have unleashed the Ryzen chip speed. Anyone else seeing this?

It's not true clock, my 5600x show 6ghz instead of manual 4.5ghz and i tested multicore benchmark no difference result betwen old 4.5 ghz and new 6 ghz, its false clock.
If you run all core at there max you can see all core down to the true speed/mhz, its up only when is in iddle.

[phoronix_is_awesome]

Hey Guys,

Just registered on phoronix just to post a feedback on kernel 5.11.

Have you guys tried to use 5.11 on Ryzen APUs? I am running 5.11 on a HP Ryzen 4700U laptop, and the thing boosts clockrates too aggressively, that I reach 103+C temperature within 10 seconds, and it is simply not safe. Also glmark2 dipped to 300-400, when it was 9000 on 5.10.15 kernel. GPU just got a 90% regression because CPU boosted too hard to reach 105C.

I don't think the frequency in-variance stuff is actually beneficial to AMD CPU users. I mean phoronix focused on the 5-6% performance boost.(at what power consumption boost cost?) That's only on EPYC and Ryzen CPU only. Not APUs. If you boost too aggressively on the CPU, the GPU on the APU is starved of power. The 4700U only got 15W to use(it averages 15W, and boosts to 25W). The laptop was running 95C full load on kernel 5.10.15, now it constantly boosts to 103-105C while the GPU takes a 90% hit.

Something is wrong here. I mean, Intel's got frequency variance since kernel 5.2 and is it why Intel CPUs now have 225W PL2 on a 95W TDP cpu? Seeking aggressive CPU boost for single threaded performance is a bad tradeoff in my mind. Especially letting a 15W U-class Ryzen running at 105C on a laptop, which is out of specs. Also people have noticed 6Ghz+ max "fake" clocks just so that freq_next = 1.25 * freq_max boosts harder.
This is simply irresponsible engineering. Too aggressive single threaded boost.

I am back to 5.10.15 for the time being. Since it is the last known good kernel that compiles zfs-dkms against zfs 2.0.3.

[reba]

How is that a problem I caused though? All I did was attempt to install the packages on an LTS 20.04 release. If I change only one thing and that's reboot to the 5.4 kernel, it builds and installs fine.

Works perfectly here:

somebody@debian:~$ uname -a
Linux debian 5.10.16-xanmod1 #0~git20210213.cea7cdd SMP PREEMPT Sat Feb 13 15:49:17 UTC 2021 x86_64 GNU/Linux

somebody@debian:~$ vboxmanage --version
6.1.18_Debianr142142

somebody@debian:~$ lsmod | grep vboxdrv
vboxdrv 536576 4 vboxnetadp,vboxnetflt

Never had a problem in the last weeks and I fire up several VMs every day.

Looks to me like you're in denial but so far everything points out at a problem on your side.


Edit: I'm using standard .deb packages and do everything with apt. All packages (kernel, virtualbox, etc) are straight out of Debian sid.
I currently use a kernel-deb from xanmod but the standard kernel of Debian works too, of course.


Edit: Are you on *buntu? Is there a strong reason for that? Otherwise I suggest going back upstream to Debian (testing or unstable); *buntu is - IMHO - a duct-tape-mess.

[perpetually high]

You’re right, reba. I’m in “denial.” This is all a conspiracy by me to create FUD and spread fake news. You guys got me. The jig is up.

edit: The real enemy here is VirtualBox, then Ubuntu, let's be clear. I'll update this thread as soon as an update fixes it. For what it's worth, I also tried it on 5.11 and no go.. it is not an OS-dependent issue, but a kernel issue that VirtualBox doesn't know how to handle yet.

[matthewsha]

It's not true clock, my 5600x show 6ghz instead of manual 4.5ghz and i tested multicore benchmark no difference result betwen old 4.5 ghz and new 6 ghz, its false clock.
If you run all core at there max you can see all core down to the true speed/mhz, its up only when is in iddle.

Thanks, I was wondering, as the temps were unchanged. Running hard all the cores do settle, but i figured that was needed to avoid high temps, and thought single cores might be more boosted. I haven't tried any benchmarks.

[reba]

You’re right, reba. I’m in “denial.” This is all a conspiracy by me to create FUD and spread fake news. You guys got me. The jig is up.

edit: The real enemy here is VirtualBox, then Ubuntu, let's be clear. I'll update this thread as soon as an update fixes it. For what it's worth, I also tried it on 5.11 and no go.. it is not an OS-dependent issue, but a kernel issue that VirtualBox doesn't know how to handle yet.

Here Virtualbox does not compile on xanmod's kernel 5.11.0, too. But the kernel is from xanmod and not from Debian, whereas the Debian's kernel is 5.10.13 and Debian's Virtualbox both are from the Debian sid repositories and work together.

I expect as soon as Debian repos are updated to kernel 5.11.0, the VirtualBox pendant will soon catch up (or be ready).

Mixing and matching is tricky and does not always work.
The xanmod kernel 5.10.16 is more recent than Debian's kernel 5.10.13 but BOTH work with Debian's VirtualBox, which is not guaranteed.
However I would not depend on this as Debian's VirtualBox package does not know about the xanmod's kernel 5.10.16 package and thus cannot be tested together.

But for now, Debian's kernel 5.10.13 with Debian's VirtualBox works.
xanmod's kernel 5.11.0 I esteem somewhat too fresh for Debian's VirtualBox.

Nevertheless the bottomline is: my distribution ensures its latest kernel and its latest VirtualBox offered by them work together.
Thus I consider yours a packaging (or "OS" and therefore Ubuntu) problem or a problem of the admin pairing incompatible packages.


Edit: Anyway, this is going too far - what I wanted to say was just another data point:
"Works for me out of the box, no problems whatsoever. That's how I do it. Why it isn't working for you? I have no idea but maybe it's this-or-that."

[perpetually high]

reba, I'll eat the crow if I'm wrong but I'm telling you - it's a kernel issue. I didn't just start using VirtualBox yesterday.

I get a lot of shit on here for running Ubuntu like a noob, but I'm well versed and have been using Linux daily since 2014. I don't get bored anymore and distro hop or any of that, and if I ever want to I could spin one up in VirtualBox (insert joke here).

One of the reasons I didn't jump from 20.04 LTS (04/20 release, no affiliation to me) is because I'm happy with where the OS is with GNOME and everything else, and I now favor long term support over bleeding edge. To each their own.

So when 5.10 got released, I chalked it up to okay, it just came out. Give them some time.. well it's been a while now and I know they're smart enough to figure it out so it begs the question what's the hold up? 5.10 is LTS as well. No 5.11 support either, so who knows when that's coming.

Again, this is small potatoes and VirtualBox doesn't owe me anything. But the fact reminds I'm stuck to 5.4 (which is a fantastic kernel for what its worth) and I compile my own kernels so I add in all the patches and good stuff so I am in no rush. I'm chillin for now, but I felt it was worth bringing up when this thread was first posted.

[Linuxxx]

Hey Guys,

Just registered on phoronix just to post a feedback on kernel 5.11.

Have you guys tried to use 5.11 on Ryzen APUs? I am running 5.11 on a HP Ryzen 4700U laptop, and the thing boosts clockrates too aggressively, that I reach 103+C temperature within 10 seconds, and it is simply not safe. Also glmark2 dipped to 300-400, when it was 9000 on 5.10.15 kernel. GPU just got a 90% regression because CPU boosted too hard to reach 105C.

I don't think the frequency in-variance stuff is actually beneficial to AMD CPU users. I mean phoronix focused on the 5-6% performance boost.(at what power consumption boost cost?) That's only on EPYC and Ryzen CPU only. Not APUs. If you boost too aggressively on the CPU, the GPU on the APU is starved of power. The 4700U only got 15W to use(it averages 15W, and boosts to 25W). The laptop was running 95C full load on kernel 5.10.15, now it constantly boosts to 103-105C while the GPU takes a 90% hit.

Something is wrong here. I mean, Intel's got frequency variance since kernel 5.2 and is it why Intel CPUs now have 225W PL2 on a 95W TDP cpu? Seeking aggressive CPU boost for single threaded performance is a bad tradeoff in my mind. Especially letting a 15W U-class Ryzen running at 105C on a laptop, which is out of specs. Also people have noticed 6Ghz+ max "fake" clocks just so that freq_next = 1.25 * freq_max boosts harder.
This is simply irresponsible engineering. Too aggressive single threaded boost.

I am back to 5.10.15 for the time being. Since it is the last known good kernel that compiles zfs-dkms against zfs 2.0.3.

I assume you are actually utilising the schedutil governor and not the performance one, right?

If so, try to play around with the value of:

/sys/devices/system/cpu/cpufreq/schedutil/rate_limit_us

Since this is the only tunable that schedutil offers for easy configuration, this is pretty much the only option one has going forward when settling on the default CPU governor.

[reba]

reba, I'll eat the crow if I'm wrong but I'm telling you - it's a kernel issue. I didn't just start using VirtualBox yesterday.

I get a lot of shit on here for running Ubuntu like a noob, but I'm well versed and have been using Linux daily since 2014. I don't get bored anymore and distro hop or any of that, and if I ever want to I could spin one up in VirtualBox (insert joke here).

One of the reasons I didn't jump from 20.04 LTS (04/20 release, no affiliation to me) is because I'm happy with where the OS is with GNOME and everything else, and I now favor long term support over bleeding edge. To each their own.

So when 5.10 got released, I chalked it up to okay, it just came out. Give them some time.. well it's been a while now and I know they're smart enough to figure it out so it begs the question what's the hold up? 5.10 is LTS as well. No 5.11 support either, so who knows when that's coming.

Again, this is small potatoes and VirtualBox doesn't owe me anything. But the fact reminds I'm stuck to 5.4 (which is a fantastic kernel for what its worth) and I compile my own kernels so I add in all the patches and good stuff so I am in no rush. I'm chillin for now, but I felt it was worth bringing up when this thread was first posted.

No crow eating necessary in my book.

As a Debian user I also get flak ("SJW", "Why not Gentoo, Arch or NixOS? They're so much nerdier" - yes they are, I'm just too dumb and too lazy) and as a vivid fan of KDE even more...

Debian (sid) is for me the sweet spot between "bloody edge enough" and "it just works". I don't want to play the sysadmin simulator.
Of course KDE 5.21 would be nice to have NOW, but it will come, eventually.
Meanwhile everything still works flawlessy and the installation really had to take some sucker punches from me.

I'm a desktop user. Therefore I expect things to work with minimal intervention. That's delivered.
Of course I can compile a kernel and tweak some options (however using frosted food and adding some salt is not cooking for me), yes, I can configure GRUB and have automated backups running with Borg - but I'm just a pleb user and enjoy using LibreOffice, Firefox and VirtualBox. No preparation needed.
Some years ago everything worked right away as I cleansed Windows 7 from this laptop and installed a Debian: Wifi, Bluetooth, volume, screen brightness, even keyboard brightness and the card ready - everything! Even the Intel iGPU and the AMD dGPU coexist peacefully. The dGPU actually completely powers down (not just idle) when not in use. I still can't believe that. What did I have to configure? Nada. Nothing. That just won me over. Two weeks later GRUB just selects the kernel, not the OS anymore.

I hope you get your VirtualBox issues fixed and please keep me informed how it turned out.
Cheers

[Cybmax]

reba, I'll eat the crow if I'm wrong but I'm telling you - it's a kernel issue. I didn't just start using VirtualBox yesterday.

I get a lot of shit on here for running Ubuntu like a noob, but I'm well versed and have been using Linux daily since 2014. I don't get bored anymore and distro hop or any of that, and if I ever want to I could spin one up in VirtualBox (insert joke here).

One of the reasons I didn't jump from 20.04 LTS (04/20 release, no affiliation to me) is because I'm happy with where the OS is with GNOME and everything else, and I now favor long term support over bleeding edge. To each their own.

So when 5.10 got released, I chalked it up to okay, it just came out. Give them some time.. well it's been a while now and I know they're smart enough to figure it out so it begs the question what's the hold up? 5.10 is LTS as well. No 5.11 support either, so who knows when that's coming.

Again, this is small potatoes and VirtualBox doesn't owe me anything. But the fact reminds I'm stuck to 5.4 (which is a fantastic kernel for what its worth) and I compile my own kernels so I add in all the patches and good stuff so I am in no rush. I'm chillin for now, but I felt it was worth bringing up when this thread was first posted.

I am running Ubuntu 20.04 with my custom compiled 5.10.16 kernel, and i know i had problems with Virtualbox when 5.10 was launched, but have no problems now. Not at home just now, but i use virtualbox.org .deb packages as explained here https://www.virtualbox.org/wiki/Linux_Downloads

deb [arch=amd64] https://download.virtualbox.org/virtualbox/debian focal contrib

(But i can doublecheck this if you want when i get home). It DOES work with Ubuntu 20.04 w/kernel 5.10.x, so that is the reason ppl are saying its "your fault".