Best way to set computing preferences for Dell 5488 All-in-Ones

mshap
mshap
Joined: 22 Apr 18
Posts: 12
Credit: 8,541,840
RAC: 13,125
Topic 214880

I recently joined the Einstein@Home project. I am using two Dell 5488 All-in-One computers with the following specs:

GenuineIntel
Intel(R) Core(TM) i7-7700T CPU @ 2.90GHz [Family 6 Model 158 Stepping 9]
(8 processors) INTEL Intel(R) HD Graphics 630 (4874MB)

Because the All-in-Ones don't have great ventilation, I have to limit the amount of computation to maintain safe temperature levels.

So far I have done this by using the BOINC computation options, setting the computing preferences to "Use at most 100% of the CPUs" and "Use at most 40% of the CPU time"

This keeps the computers at a reasonable temperature, but causes some pulsating fan noise. Looking at the task manager I find that the percentage of CPU time actually varies between 0 and 100% in a repetitive pattern, with the average being about 40%. The fan noise follows the same pattern as the pulsating CPU usage.

My question is this - is there a better way to set the computing preference that would achieve the same level of thermal management with less fan noise?

Thanks for your help.

Mark

Zalster
Zalster
Joined: 26 Nov 13
Posts: 2,976
Credit: 3,212,011,146
RAC: 1,943,176

I would not suggest running

I would not suggest running all 8 of your threads unless you have water cooling for that CPU.  

Probably better to set "Use at most 50% of the CPUs" that should help decrease the temps on the CPU and might lengthen the life.   

If you want a better idea of what your cpu is doing, I recommend SIV64X.  It gives you a better idea of how your computer is doing in general, not related to work units.

 

http://rh-software.com/

mshap
mshap
Joined: 22 Apr 18
Posts: 12
Credit: 8,541,840
RAC: 13,125

I tried reducing the number

I tried reducing the number of CPU's being used. Changed the setting first to "Use at most 50% of CPUs" with various settings of "Use at most xx% of the CPU time." That did not affect the fan pulsating behavior much.

Then I tried "Use at most 25% of the CPUs" with various settings of the "Use at most xx% of the CPU time." I found that with "Use at most 100% of the CPU time" the fan pulsing stopped and the overall CPU usage leveled out at around 32%. The fan no longer pulsed, but was running at a fairly high speed, and the air temperature at the vent was higher than with most other settings.

I ended up going back to "Use at most 100% of the CPUs" and "Use at most 30% of the CPU time." The fan still pulses but overall is fairly quiet and the temperature of the air coming out of the vent is much lower than when only one CPU (2 threads) was in use.

The interesting thing is that on my second "identical" computer, I can run with "Use at most 100% of the CPUs" and "Use at most 40% of the CPU time" for the same noise level and vent temperature.

Mark

 

Betreger
Betreger
Joined: 25 Feb 05
Posts: 812
Credit: 439,338,735
RAC: 123,661

Try opening up the computers

Try opening up the computers and cleaning out the accumulated dust bunnies.

mshap
mshap
Joined: 22 Apr 18
Posts: 12
Credit: 8,541,840
RAC: 13,125

That's a good suggestion. I

That's a good suggestion. I will have to read the manual carefully before doing that, because they are all-in-ones. The procedure for opening them up is a bit tricky.

 

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 4,927
Credit: 30,513,295,783
RAC: 34,476,307

mshap wrote:My question is

mshap wrote:
My question is this - is there a better way to set the computing preference that would achieve the same level of thermal management with less fan noise?

Hi Mark,
Welcome to Einstein@Home.  Thanks for participating in the project.

I'm not familiar with your particular hardware.  I presume it's something like an iMac where everything is in the back of the screen itself.  I'm not sure what OS you are running as your computers are hidden.  I presume it's Windows.

I ask because I became aware of something a while ago that may be of interest to you and perhaps will explain why running a heavily reduced number of tasks doesn't have more of a cooling benefit for you.  I run Linux with both CPU cores and GPUs doing the crunching.  I spent a bit of time adjusting the number of active CPU cores since the GPU was much more efficient in producing results.  The intention was to maintain pretty much the same output (slight reduction only) but actually make things more efficient by reducing the power consumption and hence the waste heat.

I experimented with progressively increasing the number of CPU cores doing crunching whilst measuring the power consumption at the wall.  I was puzzled by what I saw.   The major portion of the power increase happened with the very first CPU core allowed to crunch.  After that, adding a 2nd or 3rd core to the mix added only very slightly to the power consumed.

I did a bit of searching around to find out why and came up with an explanation that seems believable.  Essentially, with multicore CPUs, it's the operating frequency that has the biggest impact on power consumption.  If all cores are at idle load, the frequency can drop right down - as will the power consumption.   With Linux, if just one core is subjected to a high load, the usual default behaviour is to ramp up the frequency on all cores - unless you change settings buried deep in the OS to do otherwise.  So it was the ramping up of frequency on all cores when only one was loaded that was causing the behaviour.  In Linux, this can be changed so that unloaded cores don't ramp up in frequency until they actually receive a load.

In Windows, there might be a similar ability to tweak this behaviour.  I know nothing about Windows so have no idea how you would go about that.  Maybe one of the Windows users here might have come across this.  In theory, your best way of controlling heat would be to use a reduced number of cores provided you can keep the other unused cores essentially at idle frequency for most of the time.  A machine being used for casual 'office' type duties is essentially at idle.

If you are actually using Linux, I could go dig out the information I found previously.

 

 

Cheers,
Gary.

mshap
mshap
Joined: 22 Apr 18
Posts: 12
Credit: 8,541,840
RAC: 13,125

Thanks for your reply Gary. I

Thanks for your reply Gary. I have two Dell all-in-ones enrolled in the project. They look similar to the i-Macs. The operating system on both is the latest version of Windows 10. The details of the system are:

GenuineIntel Intel(R) Core(TM) i7-7700T CPU @ 2.90GHz [Family 6 Model 158 Stepping 9]

Number of processors: 8
Coprocessors: INTEL Intel(R) HD Graphics 630 (4874MB)
Operating system: Microsoft Windows 10 Core x64 Edition, (10.00.17134.00)
 
The system's strength is in the CPUs, and its weakness is in the GPUs. I think the project knows that, because I get mostly continuous gravitational wave search tasks to process with a few gamma-ray pulsar search tasks thrown in.
 
I agree with your comments about the CPU loading. The heat really went up when I tried running on just 1 of the 4 CPUs at 100% of the available time. I seem to get the best performance by using all 4 CPUs and cutting back on the duty cycle (35% on one machine and 40% on the other).
 
The internal fan in each computer ejects air through a wide slot at the top of the computer, so it's easy to check the temperature of the exhaust air. I ended up trying a number of combinations of CPUs and duty cycles. The lowest ejected air temperature was obtained by using all the CPUs with a low duty cycle. There still is some fan noise, but it's not objectionable.
 
I checked the manual for these machines, and the fan is connected to a heat sink that is connected to the motherboard. The fan actually is located beside the motherboard. The mechanical configuration is similar to a laptop. I'm guessing that keeping all the CPUs working at a reduced duty cycle results in more efficient use of the heat sink.
 
Mark
Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 4,927
Credit: 30,513,295,783
RAC: 34,476,307

My last experience with

My last experience with Windows was in about 2006 with Windows XP.  I do see people posting from time to time about various Windows utilities that can monitor/adjust parameters to do with frequencies/CPU loads/core temperatures, etc.  If I were in your shoes, I'd be investigating various third party utilities to see if any could lock tasks to particular cores and monitor/adjust the frequencies of the other unused cores to use an idle frequency.  Since all the cores are in the one package, the overall package temperature should be more stable and less likely to cause fan speed to cycle up and down.

The thing I don't particularly like about a duty cycle is the rapid (and violent) swings in internal core temperatures that must be happening.  I imagine the fan sensor doesn't really see a violent swing since it's probably looking at a much dampened package temperature.  This is all just conjecture on my part but I vaguely remember reading a discussion a long time ago about potentially harmful effects of rapid thermal cycling deep in CPU cores.  I had the impression that it was beneficial to avoid rapid load cycling if possible.  Maybe, someone who knows more about this might be able to comment.

 

Cheers,
Gary.

Richie
Richie
Joined: 7 Mar 14
Posts: 460
Credit: 1,548,552,667
RAC: 911,284

If you want to experiment

If you want to experiment even more there's this software called TThrottle:

https://efmer.com/download-tthrottle/

You can set max temps for Core and GPU. Then this program will handle the limiting. TThrottle will recognize when Boinc is running and what tasks are running.

I hadn't tried this software in a while, but I'm running it on Windows 10 (with AMD GPU) right now to see if it is stable. Seems to work well so far. Ps. I have currently set max temp of R9 390 to for example 73C and it stays there. These tasks that run under limiting have validated without problems so far. I was worried about that, but no errors or other problems so far.

I recall keeping the Graph enabled and graphs tab open was causing problems in the past (sometimes caused my computer to freeze after a while), so I unchecked that option this time. I also disabled the additional temp window.

TThrottle doesn't support Intel GPU, but it would still let you test how things would work if you set processor time (and CPU cores?) to  100% in Boinc... and then just limit max temps with TThrottle. You can actually see the temps and limit them in real time. That might help to see how the twins there are handling temps while crunching.

This additional software on itself will use some CPU cycles, but for me, that is unimportant. The whole purpose and goal is different. I won't be running anything on 100% anyway. Being able to crunch anything at all is more important for me, as early summer is already raising up temperatures around me. So, in addition to what choices Boinc already gives, I see this software might give additional choices how to keep crunching even at daytime.

I'm not running it under heavy load, so I can't say if it would remain stable in extremely busy scenarios.

Betreger
Betreger
Joined: 25 Feb 05
Posts: 812
Credit: 439,338,735
RAC: 123,661

Until you open those things

Until you open those things up and clean out the dust all these suggestions are very well putting a band aid over a hardware problem.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.