zenpower VS ryzen_smu

had same issue as well, had to install zenpower

Not sure if it works the same on a Steam Deck, but for desktop Linux, I use zenpower - https://github.com/ocerman/zenpower

Put in a bug/feature request on zenpower

From what i could gather the default k10temp might not see power draw. Arch wiki recommends installing zenpower, here is link with instructions for Ubuntu. CPU frequency I have no idea, it should work OOTB.

Lastly, I found this thing called zenpower. Installed the module, and loaded it, but lm_sensors just does not seem to even see it...

That said, there is Zenpower and Ryzen_SMU, but they're one-man-band projects, and I don't think they have any official help from AMD.

You can get more sensors that are exposed via the SMU (like SVI2 TFN) via zenpower and ryzen_smu.

Actually that’s precisely how I made my linux monitor for Zen 2/3. Zen 4 wouldn’t be any different.

I still wondering how to control CPU, found many stuff such as amd_pstate, ryzen_smu, RyzenAdj, amdctl. Now I'm trying to figure something out but it has too many options ☺

P.S. Since you’re using a 3700X, you can check out my project that I wrote for my own 3700X here that allows you to monitor the processor in far more detail than Linux is capable of showing. You’ll see per-core temperatures, true effective frequency per core, true core voltages (per core) and tons of other info. May be interesting to you so that’s why I mentioned it.

The CPU calculates effective clocks itself. It does this by using the PLL frequency which it constantly keeps track of and is able to calculate exactly how many cycles the processor was not gated for (C1/C6). All of this is done in a coprocessor, the SMU, so there’s no variations due to polling inaccuracies, it’s not an average — it is the actual PLL frequency that excludes gated cycles. HWiNFO64 does not show this true value but rather attempts to calculate itself via polling which causes an ‘averaged’ value.

For Linux I made my own version that pulls effective frequencies directly from the SMU, so you don’t run into ‘averaging errors’ due to kernel timing weirdness as this is calculated off-chip on the cIOD of the processor independently of cores.

Source: I literally reverse engineered Ryzen Master to port it to Linux. You can see my project and it’s source code. That shows everything Ryzen Master does (look in userspace/monitor_cpu.c).

How can I state all this confidently? Because I've reverse-engineered Ryzen Master myself and even wrote a Linux driver to show you exactly what Ryzen Master shows. In fact there's a lot of information it leaves out as well. I haven't ported it to other processors aside from my own 3700X but this spinoff project does add support for the 5000 series and quite a few other 3000 series processors. If you're looking for a "Ryzen Master" like monitoring tool, that will almost certainly work.

As the person who ported SMU control based power metrics to Linux and indirectly Zen 3 power monitoring, I’ve received at least a dozen 5600X power metrics samples and all of them listed, at least as reported via the SMU, the same thing Ryzen Master pulls its info from, PPT as 76 W. AGESA versions ranged from 1.1.06 (iirc) all the way to the current 1.2.0.2. The actual fuses themselves report a baked in 76 W limit.

BTW, this is terrific!

Sleeping cores in Ryzen Master actually indicates whether a core was in a C0 state for less than 7% of the time, seen here from a basic reverse-engineering of Ryzen Master.