Folding Forum

Interesting thread. I have reluctantly stopped folding on my CPUs as I just cannot seem to get the temperatures under control. I have two systems that are available for work 24x7:

New AMD 3950X + RTX2070 Super
Older Intel I7-4770K + GTX1650

On both the GPUs run comfortably, the 2070, especially, getting through a commendable amount of high value work. However I configure the CPUs I still see peak temperatures up in the 90s and that's uncomfortably high.

On the 3950 I have even tried reducing down to 8 CPUs (of 32) to no avail. I'm using a AIO water cooler and that certainly pulls the heat away but the temperatures on the CPU jump up very quickly and it seems like the cooling system's thermal lag is too great. I need to experiment with this some more but for now it's just the two GPUs folding.

Is there any OC on those CPUs. I know they can run hot but 90 is completely atypical in normal use. Usually your fans will ramp up and cool it down way before that. Even watercooling will hit a transfer limit.
My current loop for example run at 33C or less but the CPU package still hit the 70s. The IHS just can't transfer the heat any faster. Even the DH15 or CF14 can't pull CPU temps down faster either.
Given no overclocking on the CPU side, I'd have to think there's really bad airflow (possibly with a small case) and more fans might help or bad thermal paste application. For thermal paste that's not conductive using extra to make sure complete coverage is ok. But I'd seriously consider that the heatsink isn't properly making contact. I'd definitely make sure the heatsink is mounted 100% properly. A bad mounting can cause high temps as the pressure between heatsink and IHS isn't sufficient.

IMHO, I don't think another air cooler will make a difference.

Thanks for the reply ipkh. Neither of the CPUs are being overclocked. The old 4770 just has the original stock air cooling fan and I could certainly improve on that. I think you're right that there might be a problem with the AIO water cooler seating on the 3950 - I shall investigate when the current GPU folding job has completed. The AIO radiator fans do ramp up as you'd expect but the CPU temperature goes up very quickly indeed, within a second or two, so I suspect the thermal connection between the CPU and the cooling system is bad.

John

I took a look and couldn't see anything obviously wrong, so cleaned up the old thermal paste and tried again with a real thin layer. It's better, slightly: average temperatures in the low to mid 80s while folding but I still see short term excursions above 90. Might have to have a rethink on this - the CoolerMaster ML120L might just not be up to the folding job.

If you're talking about the 3950X, AMD just runs hotter to get good performance. Maybe try with out the PBO stuff. CPUs can deal with high temps and AMD just seems more willing to let things get hot.
Intel doesn't tend to do that. But Intel still rates their chips at 100C before throttling. They just don't regularly hit that temp unless overclocked or really bad heatsink stuff.

Thanks ipkh. I had to Google PBO (Precision Boost Overdrive for those equally in the dark) and went off on a research project in the depths of the BIOS. Ahha, there it is, set to Auto, whatever that means. I've now turned it off, re-enabled the CPU slot and am awaiting a WU with interest.

Once upon a time I thought I understood computers. I think that was some time in 1972. Almost 50 years of hardware and software tinkering later I now realise how wrong I was!

I hear you there, I've been building them for decades now and am always picking up new things and tricks.

G3WGV wrote:Thanks ipkh. I had to Google PBO (Precision Boost Overdrive for those equally in the dark) and went off on a research project in the depths of the BIOS. Ahha, there it is, set to Auto, whatever that means. I've now turned it off, re-enabled the CPU slot and am awaiting a WU with interest.

Once upon a time I thought I understood computers. I think that was some time in 1972. Almost 50 years of hardware and software tinkering later I now realise how wrong I was!

You could also throttle the CPU frequency very easily from the power management console in Windows. Please see my earlier post regarding this. It will really help with power consumption (important for my laptop battery) and heat.

Well I tried folding with PBO disabled and it made no noticeable difference. I'm looking into a different AIO water cooling system, as it seems the current one isn't working well enough. Any suggestions for suitable, preferably single fan, systems welcome. For the time being, CPU folding is disabled.

Konadreamer wrote:You could also throttle the CPU frequency very easily

Yes but this is my new software development system and the entire reason for building it was for better performance in that and other resource hungry jobs. Folding is one of those jobs, for sure, and is the reason that I specified a much more powerful GPU than I otherwise needed. It shouldn't be necessary to throttle the CPU, especially given how much I just spent to get a top flight processor!

TBH I suspect my conclusion will be that CPU folding is just too invasive on machines that have other work to do, whereas GPU folding has little impact other than increased electricity bills.

G3WGV wrote:Well I tried folding with PBO disabled and it made no noticeable difference. I'm looking into a different AIO water cooling system, as it seems the current one isn't working well enough. Any suggestions for suitable, preferably single fan, systems welcome. For the time being, CPU folding is disabled.

A few reviews:
https://www.tomshardware.com/reviews/be ... 181-2.html

https://www.pcguide.com/reviews/best-aio-water-cooler/

https://www.techradar.com/news/best-cpu-coolers

jrweiss wrote:A few reviews

Thank you. Thinking...

Interesting that CPU WUs typically attract less than one tenth of the points that similar length GPU WUs get. I suppose the points represent a reasonable proxy for the WUs perceived value to the task at hand. Perhaps, given that I have two decent GPUs chuntering away, there is little point in continuing with CPU WUs.

G3WGV wrote:...I suppose the points represent a reasonable proxy for the WUs perceived value to the task at hand. Perhaps, given that I have two decent GPUs chuntering away, there is little point in continuing with CPU WUs.

While it is true that the PPD reflects the scientific contribution, do note that some projects are CPU only while others are GPU only. Reason is that some simulation/features are CPU only and thus, while you may not get that many points on your CPU, you are still doing valid scientific work that the GPU can't do

G3WGV wrote:Well I tried folding with PBO disabled and it made no noticeable difference. I'm looking into a different AIO water cooling system, as it seems the current one isn't working well enough. Any suggestions for suitable, preferably single fan, systems welcome. For the time being, CPU folding is disabled.

Konadreamer wrote:You could also throttle the CPU frequency very easily

Yes but this is my new software development system and the entire reason for building it was for better performance in that and other resource hungry jobs. Folding is one of those jobs, for sure, and is the reason that I specified a much more powerful GPU than I otherwise needed. It shouldn't be necessary to throttle the CPU, especially given how much I just spent to get a top flight processor!

TBH I suspect my conclusion will be that CPU folding is just too invasive on machines that have other work to do, whereas GPU folding has little impact other than increased electricity bills.

It's due to the core placement within the SoC you and ever other 3rd Gen Ryzen user has, it's not bad, but the mobo oems, cooler oems, even Amd, could have realigned the cooler hold down brackets to better center the cooler's thermal plate over the chiplets.
https://youtu.be/JlJDv68fEcM

Look at that video.

Edited for typo

G3WGV wrote:Interesting that CPU WUs typically attract less than one tenth of the points that similar length GPU WUs get. I suppose the points represent a reasonable proxy for the WUs perceived value to the task at hand. Perhaps, given that I have two decent GPUs chuntering away, there is little point in continuing with CPU WUs.

That is a viable option.

However, I have noticed that current CPU WUs for the Covid project are returning 50-75% the PPD of the GPU WUs on my 1050ti. Looks like they have a significant priority!

Remember that F@H is a VOLUNTEER project. Contribute to the level you feel comfortable. IF an AIO cooler will help you in your other computer work, then consider buying one. Otherwise, feel good while you contribute with your GPUs.

FWIW, F@H is what prompted me into looking more closely at cooling air flow in my systems. Experiment and learn!

TxRedneck wrote:It's due to the core placement within the SoC

Interesting video, thanks. It certainly looks like that could be part of the problem. For the time being I have decided to try a different cooling method - instead of the AIO water cooler I will experiment with a Dark Rock Pro 4 air cooler. These seem to get excellent reviews and should have lower thermal latency than the water cooler. It's effectively a nil cost experiment for me as I can use the lesser cooler elsewhere on an upcoming project that is much less demanding. Parts ordered!

TxRedneck wrote:However, I have noticed that current CPU WUs for the Covid project are returning 50-75% the PPD of the GPU WUs on my 1050ti.

I have noticed this too. Although the points per WU are lower, there are more WUs per day and they tend to be shorter. And I do very much want to be contributing to the Covid research, which is largely CPU based.

jrweiss wrote:Experiment and learn!

Very much so! F@H has certainly shaped my thinking, for example the decision to get a much more powerful GPU than my usual computing activities demand. The cooling issue has caught me somewhat unawares though - I've never really had any trouble before but that's all changed, partly because of the much higher CPU loading but no doubt exacerbated by my decision to go AMD with its off-centre hot-spots.

Never had these problems on the DEC PDP8s/PDP11s and IBM 360s that I cut my computing teeth on!