AMD Posts New "AMD-PSTATE" CPUFreq Driver Leveraging CPPC For Better Perf-Per-Watt

[ciupenhauer]

This laptop right here idles at around 4-5 Watt (says powertop) / 1.8 Watt (says s-tui) with disabled dGPU but WiFi on, KDE and normal backlight brightness. I have pictures to prove it if necessary

I have 5600H and can't get it to idle below 6.4W. how did you get that low?!? and also, proof pls

[Nille_kungen]

Thanks for this and the explanation in the commit.

https://git.kernel.org/pub/scm/linux/kernel/git/rui/linux.git/commit/?h=amd-pstate-dev-v1&id=abfcbc164c1aa0c63d5e256854bad977a9645586

[numacross]

That's true however Zen 2/3 CPUs idle power consumption is simply horrible. My 5800X idles at ... ~21W. Hopefully AMD will address it sooner that later but I don't have much hope for Zen 3+ CPUs. Maybe in time for Zen4 AMD will finally be able to compete on this metric with the competition.

Speaking of this news: it's truly great, albeit four years late. Maybe Valve can push AMD to release the documentation for Zen CPUs MSRs as well to enable proper and full monitoring which is simply cringe worthy at the moment.

Here's what you can see in Windows:

While I agree that by its design desktop Zen CPUs are at a disadvantage due to the chiplet design, your measurements can be faulty as well. Your Power Reporting Deviation is 152%, so the package power measurements can't really be trusted. What do your VRM sensors say, if available?
The comparison to HWiNFO is also a bit problematic, because they use proprietary and NDA-ridden methodologies to obtain their data. At the same time Zen's power monitoring capability on Linux is abysmal, and it's AMD's fault.

[avem]

While I agree that by its design desktop Zen CPUs are at a disadvantage due to the chiplet design, your measurements can be faulty as well. Your Power Reporting Deviation is 152%, so the package power measurements can't really be trusted. What do your VRM sensors say, if available?
The comparison to HWiNFO is also a bit problematic, because they use proprietary and NDA-ridden methodologies to obtain their data. At the same time Zen's power monitoring capability on Linux is abysmal, and it's AMD's fault.

The Power Deviation figure should only be quantified and verified when the system is under full load. The screenshot was taken when the system was fully idle. There's no error.

[avem]

This laptop right here idles at around 4-5 Watt (says powertop) / 1.8 Watt (says s-tui) with disabled dGPU but WiFi on, KDE and normal backlight brightness. I have pictures to prove it if necessary

I was talking about desktop Ryzen 5000 CPUs. Laptop AMD CPUs are wired quite differently. Meanwhile Intel mobile CPUs idle at around 1W, so again, a lot of room for improvement.

As much as people tout the performance per watt ratio, idle power consumption is still king for absolute most people out there.

[numacross]

The Power Deviation figure should only be quantified and verified when the system is under full load. The screenshot was taken when the system was fully idle. There's no error.

You're correct, I re-read my own link
I'm still curious about what the VRM reports.

Edit: Also I see you're running your memory above the officially supported max of 3200MHz, so maybe that's the cause of SoC consuming so much power? What's the DRAM voltage? Do you have SoC/Uncore OC mode enabled since that prevents InfinityFabric from downclocking?

[reba]

I have 5600H and can't get it to idle below 6.4W. how did you get that low?!? and also, proof pls

s-tui:



That was on "normal idle" with WiFi on, etc., the CPU's around 4 to 5 W regarding powertop.
The wall-power-meter shows around 8 to 9 W with battery not discharging and not charging.
But I was not even trying.


Now behold... powertop, after the display was off for some time and quickly taking the photo:



Disabling the dGPU, switching to the powersave governor, letting the display turn off and it goes down to that shown above.
Amazing, considering it is still "full on" and not sleeping.
On an 8 Cores, 16 Threads CPU that I have seen boost to around 4,5 GHz.
I am not sure if I disabled WiFi or not.

Times do have changed since Ivy Bridge 2013...

Disclaimer: However I am not sure if these numbers are solid. They look okay but the tech is quite new. The wall-power-meter agrees, tho.

[Myownfriend]

While I agree that by its design desktop Zen CPUs are at a disadvantage due to the chiplet design

What about chiplets puts Zen CPU's at a disadvantage? As far as I can tell, chiplets allow them to more aggressively bin cores, add a bunch of cache, and increase core count in a cheap enough way that they can sell 64-core/128-thread chips for a third of the price of Intel's 28-core/56-thread chips.

[avem]

You're correct, I re-read my own link
I'm still curious about what the VRM reports.

Edit: Also I see you're running your memory above the officially supported max of 3200MHz, so maybe that's the cause of SoC consuming so much power? What's the DRAM voltage? Do you have SoC/Uncore OC mode enabled since that prevents InfinityFabric from downclocking?

1) The same idle CPU power consumption can be observed while RAM is running at default 3200MHz.
2) The only thing I've done to RAM is changing its speed and tightening some timings. Its voltage remains at 1.36V. No other electrical parameters have been changed in BIOS - everything is running by default.
3) The CPU is running 100% stock. No overclocking of anything including SoC/Uncore. In fact only RAM is "overclocked" (I put it in quotes because I don't think raising its speed amounts to OC'ing per se, more so considering the IF speed is at 1800MHz which is the figure AMD calls "standard". 1900-2000MHz amount to overcloking).
4) I've had Ryzen 7 3700X in the same motherboard. For early BIOS releases its idle power consumption was around 17W, later on it decreased to ~11W after a few optimization made to the AGESA code. Right now I'm running AGESA 1.2.0.3 Patch C. There's a new beta version of AGESA available but so far only MSI has released BIOS updates with it.
5) I've asked AMD three times (two times their support, one time ASUSTechMKTJJ) if Ryzen 5000 CPUs idle power consumption could be improved. Unfortunately none of them had any ideas.

So, that's it. I'm not terribly concerned with it but it's still kinda sad to see considering Core i9 11900K idles at around 6W while being built on a much inferior node (14nm). As far as I understand it's AMD's IO complex in desktop Ryzen 3000/5000 CPUs which doesn't properly power down. If you check the screenshot again, CPU cores themselves draw basically nothing.

[stormcrow]

how can they even target just once damn cpu? doesnt even make sense. I'm betting it's the whole line. or maybe just the H series or U series?

Are you asking technically, or are you asking rhetorically? Technically you just do an ID check on the CPU and match them against a table of known good models. That's trivial. They're looking for gotchas along the line. AMD's own engineers know just how much power states can screw up either from CPU bugs or mainboard OEM screwups.