Core temperatures

[MeeLee]

I found the easiest way to adjust the Ryzen 9 series CPUs, is to adjust the PBO.
Go into BIOS, click AMD Advanced Overclocking, Set it to ECO. Go to PBO, check the values, and lower/increase them.

Depending on your system, you may need to disable PBO.
Or, if you enable PBO, limit power and Amp draw, and temps.
If your BIOS looks anything like this:


Start with setting PPT limit to 85W.
TDC limit to 60A
and EDC limit to 90A.
Then set platform thermal throttle to 80C or 75C.
If the board still increases temps much beyond this, draw a few watts and amps from TDC and EDC.
There's no way around it, but if you lower the stock values with PBO, your CPU will run at lower frequencies.
Some people find disabling PBO entirely, helps them keep the temps at bay.

[ajm]

I read that you have "TPU". I suppose that you mean the utility that comes with Asus motherboards?





Once there, you can change the frequency ratio (on my pic: 037.00) using the left/right arrows then click "Apply". For example you can choose a level that is a bit lower than base speed. It will have the same effect than using the tweak I was talking about above (Processor performance boost mode). But it will be only temporary, until the next reboot, whereas the Regedit tweak is permanent.

That said, the real way to handle this is to install a good cooling solution. Albeit, according to my experience, cooling a Threadripper sufficiently for folding at boost speed can be exorbitant. So much so that I suspect that there is some incompatibility between FAH and the Threadripper. If so, it would be important to raise the issue, to seriously measure the thermals with FAH and with similar software, in order to push for a solution. If I'm right, and if my experience is indicative of the general situation, that incompatibility can destroy a Threadripper. My first one died after a few weeks of 24/7 folding at base speed. It might be a coincidence of course but if it isn't, it really must be addressed, no?

[PantherX]

...the real way to handle this is to install a good cooling solution. Albeit, according to my experience, cooling a Threadripper sufficiently for folding at boost speed can be exorbitant. So much so that I suspect that there is some incompatibility between FAH and the Threadripper. If so, it would be important to raise the issue, to seriously measure the thermals with FAH and with similar software, in order to push for a solution. If I'm right, and if my experience is indicative of the general situation, that incompatibility can destroy a Threadripper. My first one died after a few weeks of 24/7 folding at base speed. It might be a coincidence of course but if it isn't, it really must be addressed, no?

I understand where you're coming from and the precaution that you're taking. Generally speaking, F@H is using well-established instruction sets that AMD/Intel have supported for many years. There's nothing in the software itself that can physically damage your CPU as long as it was operating within it's specifications. Moreover, there are other donors who have Threadrippers who fold 24/7 without issues.

IMO, I would focus on specific factors within the failed system to see what could cause the issue. If F@H was causing a hardware failure, our Forums would have noticed a massive spike in reporting. That hasn't happened. Thus, see what settings the motherboard had, the temperatures, voltages, etc. If everything checks out, the option that you had a faulty CPU could be valid.

Nonetheless, it would be interesting to see what you find (or don't find) "odd" or "unique" in your system when compared to other systems.

[ajm]

Well, with the Threadripper, FAH produces thermals that are at odd with what is expected (as well as crippling 997 errors). I get higher temps running FAH with half the threads than running Linpack Extreme with all threads (see above). Granted, I hadn't tried that with my late Threadripper, but the thermals are otherwise the same. And it would be a terrible bad luck to get two faulty units one after the other. Now, I do have a third 3970x, and a TRX40 MB, and enough RAM on the side - I could build a test bench. Would 3 threadrippers behaving the same convince you?

I also ran Linpack on my two folding rigs (X299 and X79) and there, the thermals are the same than with FAH. There is something unique when folding with Threadripper. And if there is a common theme among people here using them, it's the difficulty with thermals (and the 997 errors).

[PantherX]

Well, with the Threadripper, FAH produces thermals that are at odd with what is expected (as well as crippling 997 errors)...

The 997 errors doesn't make it easy. I am a bit suprised to see a lack of mention here. Not sure about Team forums.

...I get higher temps running FAH with half the threads than running Linpack Extreme with all threads (see above). Granted, I hadn't tried that with my late Threadripper, but the thermals are otherwise the same...

Since folding is FPU intensive, using 50% of "threads" would be the same as using 100% of FPUs. There is extra heat generated by using SMT but from what I have read, it won't be that much as the FPUs are already at high utilization.

...And it would be a terrible bad luck to get two faulty units one after the other. Now, I do have a third 3970x, and a TRX40 MB, and enough RAM on the side - I could build a test bench. Would 3 threadrippers behaving the same convince you?...

Before we give into bad luck, was there anything that was common across all the CPUs? Are you using the same motherboard, same PSU, etc. if everything was 100% different without any reusing of the components and you also ruled out electrical issues, then it is something worth digging into since it is unexpected behavior. While not frequently reported, I would be interested in figuring out the root cause for this.

...There is something unique when folding with Threadripper. And if there is a common theme among people here using them, it's the difficulty with thermals (and the 997 errors).

If you're part of a team, maybe post in their forum and see if there are other reports too? Maybe post in AMD forum to see what's happening to your system? If it was a widespread issue, there would be a lot more noise about it.

[ajm]

Even if 50% is the same as 100% (which surely is not correct - I remember tweaking the temps by choosing the number of threads with several processors), FAH generates more heat than Linpack - on the Threadripper.

The motherboard is also brand new on my current system, and so is the RAM. The PSU (Be Quiet Dark Power Pro 1000 W) is the same. I could replace it. Will look into it.

I'm part of a team (182116), but their forum is not FAH intensive, at all. This here is the best choice, as I see it.

[ajm]

OK, I ordered a new PSU and it has been sent, so I should get it tomorrow and then have a complete new TRX40 system in the evening. I'll redo the tests above and we'll see.

But let's call here all Threadripper Folders: do you have thermal problems while folding? Do you fold 24/7 with your Threadripper at boost speed? If not, why not? Do you have 997 errors issues when FAHControl is open? Would you agree to test your system with Linpack Extreme (which should produce the same stress and heat as folding)?

I want to underline here that I'm NOT looking for trouble or trying to get some new free hardware (AMD already replaced my dead 3970X), nor am I suggesting in any way, shape or form that FAH software is bad for your hardware. I fold 24/7, North of 4M/day and I hope to do more. But for now I have to suspect that at least the new Threadrippers (39x0X) have some sort of incompatibility with FAH, and if so, FAH and AMD should try and solve it. That's all.

[Neil-B]

OK, so I may get pilloried for even suggesting this - apologies if I offend - and I in no way state this as actual fact, and as a non AMD user this is more idle speculation taking into account corporate/commercial style decision making:

I wonder if maybe it has nothing to do with your setup nor to do with the FAH software - but rather is simply down to choices made by AMD when designing/manufacturing current Threadrippers to meet their intended mainstream market, in that to push what they have into these relatively extreme packages they have only succeeded in doing so by accepting compromises/limitations in areas they care less about or haven't tested to extremes?

FAH software is pretty much an edge case as to how intensively it utilises CPU compute resources over extended periods - this is fairly well evidenced/accepted I think ... and some CPUs/Platform configurations manage heat better than others (with laptops at one end and bespoke rigs at the other?). AMD may simply have pushed current Threadrippers to the point where they are borderline survivable on extended compute of the type FAH does but absolutely fine for the uses that AMD are targeting these processors for? ... and the MoBo manufacturers may also have a part in this as they push BIOS configurations and setups for maximums again aimed at the intended markets but that make the matters worse when using FAHlike intense compute.

If Threadripper performance at the design stage has been maximised on the assumption of a spiky workload and has really been pushed to the limit to achieve this then the very different sustained heat generation profile from FAH might just be too much for them - or at least under certain configurations/setups.

I would agree that if this is the case then there should probably be many more failures ... but for the current crop of Threadrippers is it still early days? and may we start to see more issues over the coming months? ... I have always supported the view that as long as a CPU is used within its max operating temp then is should not be harmed - maybe for sustained use at maximum temps generated in the way FAH does this just isn't true for these CPUs

Experimentation, observations and time will tell ... and whilst I really hope it isn't a design compromise issue I wouldn't totally discount it at this stage?

and this isn't about blaming AMD, just acknowledging that maybe they optimised these CPUs for a different purpose ... Formula 1 grand prix engines would die if used for 24hr Le Mans race unless very heavily limited/monitored - perhaps Threadripper is the CPU equivalent of an F1 engine?

[ajm]

Yes, but it should show also with a software such as Linpack Extreme, no? And this is not the case, at least my present Threadripper doesn't mind Linpack's stress test and doesn't show problematic temperatures. It gets significantly hotter with FAH, though.

At least in the case of my dead Threadripper, it was not an F1 engine pushed to the limit: I used it only at base speed and never at 100%, so as to stay within the limits of with very reasonable temperatures (under 75°C on the package). All the while I was searching for a cooling solution that would allow it to deploy its whole potential. And I eventually found it (2 Alphacool Eiswand arranged in parallel and an Absolute Threadripper 3+ CPU Block) but by the time I got all that, my first 3970X had passed.

That said, it is indeed very possible that AMD innovated a bit too far, and that it reveals itself when using FAH.

[Neil-B]

Not necessarily, as I guess that Linpack and FAH work the processor differently if they give different temps? … Linpack may come within the use envelope AMD made allowances for - the nature/complexity of FAH may just push it a bit far? … and who knows, if I was a cynical person I might wonder if a CPU manufacturer might ensure that standard stress test tools where "catered for" in the same way a motor vehicle manufacturer might "cater for" emissions tests

… and before anyone gets the lawyers involved … THAT WAS A JOKE !!!

[ajm]

Not necessarily, as I guess that Linpack and FAH work the processor differently if they give different temps?

According to my tests so far, the difference in temperature appears only on the Threadripper. At least, I haven't been able to observe a heat difference between FAH and Linpack Xtreme on Intel processors (i7 and i9).

[Neil-B]

… but different architecture, manufacturing, clocks, core counts, business practices, engineering approaches, etc - so not really comparable? … Intel have tended to be relatively conservative about how far they push their clocks, cores … with Threadripper AMD have pushed boundaries quite a long way, hence why it would not surprise me if some market based compromises with the latest Threadrippers which haven't been made before by them or by Intel.

… but this is simply a thought … it may well (hopefully most likely) is totally off base and wrong

[ajm]

All those aspects will affect the absolute numbers, but not the comparison between two softwares running on the same system. I get much better temps on the custom loop with close to 3 litres of water injected by 4 pumps through 2 360 full-copper rads with 6 push-pull fans each than on the air-cooled Intel processor. But I get some 15C difference between FAH and Linpack on the Threadripper and close to none when comparing both on an i7 and an i9. It shows that there is something unique to the Threadripper that shouldn't be there.

[MeeLee]

There's been a recent article, on motherboard manufacturers that go beyond the specs of the CPU.
AMD with their Ryzen series CPUs, have been more the victim of this, while Intel chips are more or less kept to the manufacturer max specs.
If your Ryzen runs at above 4Ghz on all cores, you may want to know that there have been reports of shortened CPU life.
Most budget motherboards won't be able to get your CPU much beyond 3,8-4Ghz. But if you have a $200+ motherboard, there's a chance that your CPU will be fed too much power (which also reflects to high temps).

Anything beyond 4Ghz or 145Watts TDP, is a very high overclock (considering that the 3900x series base clock is 3,5-3,6Ghz and rated at 105W).
You might hit higher frequencies when the CPU load is all core but lighter. But at full load, most 3900x CPUs should load to 3,95Ghz (some even as low as 3,65Mhz).
CPU temps are also very confusing.
Some motherboards (Gigabyte, Asrock, Asus) doesn't show the actual CPU temps, but a sensor inside the CPU socket, showing higher than CPU temps.
It may show 95C before the CPU is throttling, but the BIOS might show the CPU temp is actually much lower (in my cases they're 80-85C); which makes this all very confusing.
As closed watercooling systems are supposed to deal with only 60C, but when the CPU hits around 90C, water cooling systems hit 60-75C.
All this makes me wonder if it's not better to disable PBO, and set the CPU frequency to 3.650Mhz to 3.925Mhz (whatever can be ran stable and cool enough).

[ajm]

AMD openly publishes the base and boost speeds of their chips, as well as their wattage. For the 3970X, for example, it's 3.7 Base, 4.5 Boost, and 280 W, respectively. Max temp is 95°C.
https://www.amd.com/en/products/cpu/amd ... pper-3970x

For the temperature, "CPU" is the temperature of the processor's socket, and "Package" is the core temperature, that is, what is inside, what really counts. They often get mixed up indeed, but using a good tool, you should always see both. Now, the CPU temp is often more practical because more stable. The core temperature on the other hand will change constantly.