Benchmarks: AMD Ryzen 7000 Series Performance Boosted With Ubuntu 23.10

[HEL88]

The AMD cores consume also more power, i.e. they're doing _something_, more than the intel cores, and yet
the results show their work is for naught, which is surprising since recent Zen have very good power management features
and perf/watt in the "usual" phoronix benchmarks is above intels.
Yes intel cores can boost better, but their power consumption is usually (i.e. seen on other benchmarks in the past)
also significantly higher than amds when they do that.

In multi-core workloads, AMD dominates, in single-core workloads, Intel prevails.

The chiplets structure is simply reflected in this - the high cost energy of SoC communication with the cores compared to the power consumed by the core itself.

[andre30correia]

Sad that even powersave is better than schedutil, but Ubuntu has another improvement waiting if they switch to performance on the next release.

Has anyone noticed that with EPP active on powersave, singlecore load doesn't raise frequencies? If I load up all cores it even gets above base clock but a single core stays below 1 GHz. Or is this buggy behavior from my platform?

my experience is, it dont make any difference in daily uses, it scales the same way, im using the powersaving with balanced_performance and have the same scores and same experience with less heat in a amd ryzen 5700u, using for light gaming, browser, video, and vms

[JacekJagosz]

I am on Linux 6.5.5 with a Renoir (Zen 2) APU, but when I run `cpupower frequency-info --governors` I get:

analyzing CPU 3:
available cpufreq governors: performance powersave

So does that mean my distribution is still defaulting to ACPI-CPUFreq? But on the other hand it is using powersafe as it should

[unwind-protect]

Is there a tool like i7z for AMD? I want to know what C-states I am in.

[Anux]

I am on Linux 6.5.5 with a Renoir (Zen 2) APU, but when I run `cpupower frequency-info --governors` I get:

So does that mean my distribution is still defaulting to ACPI-CPUFreq? But on the other hand it is using powersafe as it should

What does cpupower frequency-info say under driver? If you only have performance and powersave you probably are on some pstate driver.

Is there a tool like i7z for AMD? I want to know what C-states I am in.

powertop

[Nth_man]

if (cpu_vendor() == VNDR_INTEL)
fast_path()
else
slow_path()

In a 2008 arstechnica.com article, a VIA Nano gained significant performance after its CPUID changed to Intel:

http://arstechnica.com/hardware/reviews/2008/07/atom-nano-review.ars/6

[oleid]

In a 2008 arstechnica.com article, a VIA Nano gained significant performance after its CPUID changed to Intel:
http://arstechnica.com/hardware/revi...eview.ars/6

Classic Intel compiler move back in the days. I'm not sure what is used to compile Lightroom and friends.

[hiryu]

Maybe with 23.10 I'll be able to properly shutdown my system... I have an x670e chipset with an AMD 7950x. When I try to shutdown with 23.04, the system eventually hands for a little bit (after the proper OS shutdown fortunately), and then begins POST'ing again. Windows and FreeBSD have no problems properly powering off. It's very frustrating.

Has anyone else encountered this with this generation of AMD chips?

[Keith Myers]

I am on Linux 6.5.5 with a Renoir (Zen 2) APU, but when I run `cpupower frequency-info --governors` I get:

So does that mean my distribution is still defaulting to ACPI-CPUFreq? But on the other hand it is using powersafe as it should

I also found this surprising when I loaded up the Ubuntu 23.10 Live USB test installation. Defaults to acpi-cpufreq driver and schedutil governor?? I thought with all the articles on kernel 6.5 that the driver should default to amd_pstate for Zen 2 and greater AMD cpus with governor of powersave.

So you still have to manually force it onto the amd_pstate driver?

[Keith Myers]

Is there a tool like i7z for AMD? I want to know what C-states I am in.

Try turbostat. Works for AMD cpus also. Displays C-states.