It seems that I cannot see my CPU temperAtures using any of the utilities I know. Has anyone been able to display the temperature of a Ryzen 3800, preferably in conky?
Sorry, wasn't sure whether this was "hardware" or not.

The Ryzen 38XX chips monitor clock frequency, SoC temp, and Applied Voltage to keep the thing within sane bounds. I was watching some video (linked from AMD's site) on youtube yesterday about it. Are you overclocking it? If not, I wouldn't sweat. It won't blow up. If you are overclocking it, you're on your own…

No, I'm not overclocking it but I do want to get the performance I paid for from both my CPU and cooler. I am not seeing the "full 4.4" when running stress testst so, rather than assuming all is OK I would like to confirm what is happening.
If I bodget my cooler and thermal paste install and I'm losing potential performance I would like to know about it. I did spend money on an unlocked chip and a stupidly-big cooler to lose performance due to something I can change.

You're not, without overclocking. You're getting the safe 'this will never blow up, but we're still faster than Intel' level of performance.

Again, based on my quick scan of Ryzen chips, Linux drivers aren't written yet and may never be. I seem to recall some way of getting 2 cores on one thread if you're not multi threaded enough. If you go overclocking, you remove some/all of the safety margin that AMD built in, and your warranty is void. You should look up the chips - they're really sophisticated, and inventive (and I'm a hardware guy). AMD dumb down the explanations - you don't have to digest techno babble or gobbledygook. Look for video links on AMD's site.

So what you are trying to say is that I wasted the money I spent for mny cooler because there is no way it could perform as well as the included cooler?
AMD themselves state 4.4GHz, I am not seeing 4.4GHz, I use a better cooler than standard. Therefore AMD were lying? There's no chance my thermal paste's not covering the IHS, no chance the cooler's not attached correctly? AMD just plain lied?
Sorry, the rest of the reply about drivers and 2 cores on one thread makes no sense.
In short, I don't think you understand my question or reasons for asking it.

coming back to the topic...
What output do you get from 'sensors-detetct'?
Have you tested some of the other binaries of the lm_sensors package?
Which version of lm_sensors?

I don't think I'm saying that. I'm saying heat is monitored, and the applied voltage & cpu frequency. So if you add extra cooling, the chip will use it, and your frequency or voltage should rise accordingly.

So sensor monitoring via the linux kernel is unnecessary, as the microcode or something very low level watches it, and will do a better job. Intel never bothered doing this. If the chip overheats, it's a chip fault. And more cooling gives more potential output.

273,

You probably need to modify your grub file as per:
https://askubuntu.com/questions/1164...n-x570-chipset

Earlier discussion:
https://github.com/lm-sensors/lm-sensors/issues/134

1 members found this post helpful.

I'll have to check my version when Io get home but I'm using the default in Sid. Sensors-detect just isn't seeing anything beyong my GPU - not even CPU fan speed is detected.

sorry, I was a bit rude but my point is that, for example, my graphics card will overclock itself dependent upon cooling and I hope my CPU will also. I have a third-party "best in class" type CPU cooler so oif my CPU is hitting it's thermal throttle I'd rather it didn't.
To put it in perspective I overclocked my old FX-8370 to a turbo of 4.6 and never saw a temperature over 65C with sustained 100%, all cores with mprime. I don't want to find my CPU if let down by a bodged thermal paste application, for example. i'm not a speed freak but would like to knopw my CPU is cool and working as well as it can.

aha, thanks, not seen these in my googling. Reminded me to mention I'm running an X470 motherboard.

No offence taken. I often explain myself poorly. It's a case of "I know what I'm talking about…" Wars start that way - WW1 for instance

The cooling has to be so spot on that IIRC AMD has some proprietary cooler (not requiring thermal paste) on the job, and thermal paste is relegated to auxiliary coolers. It's not like with like when you compare different cpu temps either. The wafer gets hot and has to be cooled. Everything from the wafer to the air blown through the heatsink is one long chain of thermal resistances. The way AMD are doing it, the sensor can be on the wafer, and hopefully fairly central in each wafer. Temperature sensor designs exist employing silicon junctions as the sensing element. You're right to be concerned about maximising cooling, as that will increase efficiency of cooling, and cpu output. Let me know if you find it possible to overpower it.

Though I do have the AMD supplied cooler (with past pre-applied) I'm using a Noctua NH-D15 with Noctua's own paste (reputedly very good).
More recently I have seen cores go to over 4.5GHz (which I know is the "single-core turbo" minimum) and all cores over 4.4 so I'm no longer worried about undercooling but still eager top know how well my thermal solution is working (thermal paste aplication etc. Being a concern) if nothing else just to know it's as cool as possible so less stressed. Heck, I may even try a light overclock should I find all going well.

Handy thing then is to watch the clock frequency. You'll hardly get a look at the voltage, because it will be internal. But if it starts throttling back frequency, you'll know there's heat trouble. If you have one of those infrared thermometers, you might get readings out of that. But it would be a case of comparison - hotter/cooler. You won't get absolutes. What matters is the wafer, and AMD have a handle on that.

Another trick I found useful back in the days I was doing stuff was measuring the current on the positive line. This, again, is comparative. The trick was to put a positive probe nearer the supply, and a negative probe near the chip, and measure in millivolts. Every inch of track has a finite although very small resistance. I had/have a fancy meter that measured with a resolution of 0.1mV. Basically, in your case, the bigger the reading, the better. If it's doing, say, 15mV normally over the section of track you can access,and it drops to 10mV, that means it's throttled back frequency or turned off some cores. If it rises to 50mv, you probably have issues! You want to avoid going direct from the power supply, as loads of other chips, disks & things confuse the situation. Find track to the cpu.

They're probably doing this the optimal way - measuring voltage drop across a junction on the wafer. The higher the temperature, the lower the voltage drop. You can profile that, eliminate non-linearity, and act accordingly.

Well, it seem that somebody has been working on this and the CPU temperature now seems to appear in sensors as "Tdie", though there's another reading "Tctl" which shows exactly the same temperature so I wonder whether things are still being worked upon. "cpu_fan" is not working yet but I'm happy to mark this thread closed as it seems to be a work in progress and I can see the temperature.
More worryingly the temperatures is getting over 70C under heavy load so I'm glad I did find the temperature as I may have to reseat my cooler. There's no wonder my multi-core boost speed isn't very impressive.