AMD Ryzen 5000 Temperature Monitoring Support Sent In For Linux 5.12

[Sanjuro]

What chip are you running?

zenpower-pci-00c3
Adapter: PCI adapter
Tdie: +28.6°C (high = +95.0°C)
Tctl: +28.6°C

I have 5600x. It's good to see its working for other people.

[creative]

Sanjuro nice! Yes, it is good see. I could be mistaken but I thought I saw someone with a 5950x posted earlier as well.

[Berniyh]

weird news, I have temperature reporting in my 5600X with kernel 5.10 (and I think with 5.9 as well)

That is just because the news (as unfortunately often here) is very imprecise and incomplete.

k10temp right now (with kernel 5.10 and before) will give you basic temperature reporting, which is Tctl and Tdie.
For Zen 2, they've added TccdX temperatures that allow for monitoring each CCD and that's what was missing here (and is added for Zen 3 with the aforementioned patch).
In addition, k10temp also showed voltages and currents for the CPUs and that would be shown as well, but it was removed from k10temp.

So basically with the patch applied you get Tccd1 and – if applicable – Tccd2.

The zenpower module reports a lot more on supported CPU's.

Yes, but following the discussion there and on the hwmon list, it seems that the values are not necessarily reliable, which is why it was removed from k10temp.
zenpower does show them, but you possibly shouldn't take them for granted.

According to the comments over there it even seems the case that some Mainboards feed wrong readouts to the CPU to achieve higher clock rates (you might remember the whole burst frequency discussion with Zen 2 …).

Bottom line is: unless AMD provides proper docs or programmers to adapt the code correctly, you can't get reliable voltage/current/power readouts for Zen CPUs.
btw, the same seems to apply to hwinfo and the like as well, but nobody seems to care there, that's the difference.

AMD is acting really weird here …

[creative]

Berniyh Known.

Even if stuff is a bit off, I would settle for an approximation, rather than nothing. My feeling at least with my board is that they are relatively accurate, having used the same cooling solutions on both intel and AMD.

On Asus, which is not perfect by any means but not terrible. My board is farely mature as far as x570 goes.

Also 90c is where the processors on 5000 series stops boosting, others have correlated the same. I have observed this myself. Stuff can't be too far off I wouldn't think.

If it is, its still not Ragnarok.

[Guest]

WTF, you gotta be kidding me, this is not already in 5.11 ?
AMD cannot seriously be more crappy than this!
We always need to wait for some people to reverse engineer it from somewhere?

Yeah. This is one of the most fundamental things to monitor, and you'll always see AMD fans rushing to defend them.

"I don't care about temperature, I just chuck it in a case and stuff it in my closet"

Not having even temperature monitoring is a monumental fuckup. They would be completely crucified if this was Windows, but because it's Linux, we should be grateful and sing praises to AMD. What a joke.

[Guest]

As Linux is not an AMD product, maybe blame the lazy Linux kernel authors for not making this a priority?

As Zen 3 is not a Linux product, maybe blame AMD for not prioritising this?

Intel added support for fundamental things like this at the beginning of 2020. Tigerlake launched at the end of 2020. Even people using Ubuntu and Fedora had out of the box support.

This is crap support from AMD.

[Guest]

A little off topic, But i just bought a 5600xt and i will be getting a new mobo soon. In reflection to the temp issue can anyone recommend a mobo that all/most sensors are picked up?

Nuvoton sensors are better supported in general than ITE. Read the motherboard manual/spec sheet online, then check if the sensor chip used is mentioned anywhere (or ask somewhere). And then check if the kernel actually supports that particular chip.

For the latest gen motherboards, looks like NCT6797 and NCT6687 are popular. I believe both are supported by the latest stable kernel version.

[Guest]

Modern CPUs (almost everything released for the past 10 years) HW monitoring does not depend on a motherboard as all the sensors are built-in and polled via MSR.



What a crap statement. Read the story about how the HW monitoring of Ryzen 5000 CPUs was removed in the first place. Hint: it has nothing to do with Linux kernel developers. Actually just follow this link and get enlightened:



Dropped with the help of your favourite Linux "hater", lmao.

I don't get how voltage monitoring could damage the hardware. That driver doesn't modify voltage or current in anyway, merely exposes it as a sysfs file. It's if something else reads it and takes action where the problem lies.

Actual voltage and current changes are either done by the CPU, motherboard or both. The OS merely requests a higher performance state.

[Guest]

I wondered the same and thought maybe the gentoo kernel I used was already patched.
I have just checked and it seems the change is only about the ccd temperature.
After applying the patch to my 5.10 kernel I now also have Tccd1 on my 5800x, where I only had Tctl and Tdie before:

k10temp-pci-00c3
Adapter: PCI adapter
Tctl: +38.4°C
Tdie: +38.4°C
Tccd1: +36.8°C

Tctl is actually kind of useless. It seems to be a time averaged temperature. Tccd temperatures are the instantaneous readings that we're used to, and reflect current state. If you stress all cores/threads, you can see Tccd rapidly rising and plateauing, and then when you stop the stressful load, Tccd rapidly falls back to idle temperatures, while Tctl decreases slowly.

However, Tctl is what's used to control the fans, so they stay ramped up when they don't need to.

[Guest]

Amd just posted a lot of Linux engineer jobs so hopefully this doesn't happen in the future

I was holding on to that hope for more than a decade like a foolish and naive person. Not going to do that anymore. It's going to take drastic changes at AMD for me to have any sort of faith in them.

All of the AMD fans here, enjoy drinking the koolaid.