Heat problems and running less workers than cores

MacMagnus · 2013-07-12, 16:04

I think I have a little heat problem when running LL on 4/4 cores (the outdoor temperature is high now in the summer, causing less cooling in computer).

It is around 90 degrees Celsius.

How problemativ is it to run the computer at 90 degrees 24/7? Will it be destroyed anytime soon?

I tried running 3 workers (on the 4 cores). The result were much better temperature readings. But, when using the program "CoreTemp", I noticed that the cores shared the workload (that is, averaging about 75% load on each core). I didnt think Prime95 were designed with parallel-programmed algorithms? Does running 3/4 cores mean 25% less work done than 4/4, or is it less efficient some way?

(I suppose the workers share the cores running 4/4 also...)

kracker · 2013-07-12, 16:18

What CPU, what cooler?

90C is hot.

firejuggler · 2013-07-12, 16:27

the reason is maybe because running 4 core @ 75% is colder than 3 @ 100?

Xyzzy · 2013-07-13, 04:19

http://www.mersenneforum.org/showthr...446#post269446

TheMawn · 2013-07-13, 22:11

90C is definitely some sort of Intel processor as all AMD's I've heard of get pissed off far above 60C. Most of the Core i processors are set to throttle at 105C but I would not want to go anywhere near that.

I don't know how the helper threads work. I only encountered them by accident myself when I was testing how memory bandwidth starved my system was.

Running three workers is probably your safest bet. If we knew your processor model and frequency and your memory frequency, we could even tell you if you would expect to see much of a decrease in throughput by running three workers instead of four.

Prime95 has very little parallel programming, but what you will some times see is the computer distributing the 3 workers relatively evenly over all available cores. This isn't parallel programming. It's just the program handing out work fairly randomly between processors. This will relieve some of the heat issues too, as you will have all the cores at about the same temperature instead of having three cores ~10C warmer than the other.

If 90C is due to abnormally high outdoor temperatures then you're probably alright to keep it running. If it is due to normal temperatures for the season, you'll want to figure out some sort of solution. We've had a couple of very warm days here which brought my CPU from an average of 72C and max of 75C to an average of 80C and a max of 85C, but this was only during maybe 6 hours of the day. This didn't worry me. If it was going to be touching 80C regularly for three or four months, I would be a little bit concerned.

Having a k-series processor, I can just clock it down from my current overclocked state. Underclocking yours could help too, but this will also affect the performance of your computer across all your applications.

I've heard somewhere that your processor's life expectancy is roughly halved for every 10C. Thermal throttling occurs at 105C so Intel is probably telling you that 100C is "okay", but this is "okay" as far as the warranty is concerned. If your warranty is good for 12 months, then your processor may last 24 months at 90C, 48 months at 80C, etc.. Bear in mind this is a very rough rule of thumb.

kracker · 2013-07-13, 23:06

In general I don't let my cpu go above ~60-70C, over that it makes me really uncomfortable.

And not defending AMD at all here, but I have run 24 hours of P95 stress test on a AMD quad at the mid 70's without a problem.

EDIT2: My Sandy Bridge(mobile) throttles down when it gets to 100C.

chalsall · 2013-07-13, 23:26

Quote:

Originally Posted by kracker

My Sandy Bridge(mobile) throttles down when it gets to 100C.

The boiling point of pure water at sea level?

Seems reasonable to me.

ewmayer · 2013-07-13, 23:37

@TheMawn: Prime95 is fully multithread-capable, but generally 1-single-threaded-job-per-core is slightly better in terms of overall throughput, thus that is the recommended option for vanilla GIMPS work.

Quote:

Originally Posted by kracker

In general I don't let my cpu go above ~60-70C, over that it makes me really uncomfortable.

My macbook Core2 CPU (running my Mlucas code 2-threaded across both cores) often gets above 80C, but yeah, 90ish and higher is worrisome.

Quote:

And not defending AMD at all here, but I have run 24 hours of P95 stress test on a AMD quad at the mid 70's without a problem.

To be fair, how much Prime95 throughput (inverse of per-iteration time at same FFT length) are you getting on AMD compared to Intel?

There has been much recent discussion on CPU cooling in the Hardware "Haswell preview" thread (see discussion starting around post #169) - couple of key takeaways:

o Intel, apparently in a silly effort to squeeze maximum revenue out of CPU sales, moved from a fluxless liquid-metal solder which they own the patent to(!) to less-efficient "paste"-style with Ivy Bridge. This is with respect to the "internal" thermal interface material, enclosed inside the chip packaging and used to join the chip's integrated heat spreader to the CPU die. This is significantly more important for overall CPU cooling than the "external" thermal interface between the chip packaging and the removable heat sink, which is the only one end users have ready access to. A related article from Tomshardware.com shows how to actually open the chip packaging (with lots of cautions) and replace the internal TIM with better stuff.

o The thread also has links to a new kind of retail "liquid metal" TIM, which can make all the difference for hot-running chips which have lesser quality internal or external TIM. The newest Haswell CPUs are gonna run very hot indeed once George finishes coding to make use of the new fused mul-add instructions. One poster to the above thread shows how running Linpack - which already has FMA support - leads to CPU throttling even though he is running at stock speed, no OCing.

kladner · 2013-07-14, 02:32

I'm not happy if my AMD FX gets close to 60 C. I know Piledrivers are less temperature sensitive than the Thuban I had before. But Core i-x chips are still more tolerant.

It seems that Intel might have decided to use up some of that inherent margin by switching to a cheaper, less effective die-to-heat-spreader interface material. It is easy to imagine that adhesive, non-molten-metal compounds are easier to apply acceptably than solder. I would hazard that material cost is less significant than manufacturing cost in this case.

It will be interesting to see if the solder interface gets reintroduced in some supreme-ultra-premium-clocked (priced) CPUs. As pointed out earlier, the reduced die size of recent Intel chips already makes the interface a choke point.

That said, the "delidding" procedure seems all too risky, with a small misstep yielding a dead chip. And, getting evenly distributed force of the right amount is harder on a smaller, more fragile target.

TheMawn · 2013-07-14, 18:24

The de-lidding process is something I would be interested in testing out myself at some point. Once the liquid metal TIMs start catching on, a person will see some MAJOR improvements in getting away from the current crap paste --> IHS --> more crap paste --> stock heatsink.

Once the processor has been delidded, a bunch of people have been applying their heatsinks directly to the die itself instead of putting the lid back on. I'd be in this boat myself. With liquid metal TIM and a decent waterblock, we might be seeing third and fourth gen chips pushing 1.5V-1.6V.

Will still need to wait for quad channel IB-E processors or for DDR4 / GDDR5 memory to take advantage of this though. 4.6GHz i5-3570k is bandwidth limited by 2400MHz memory, so there isn't much point in trying to push 6.0GHz on the same memory.

kracker · 2013-07-14, 21:29

Quote:

Originally Posted by TheMawn

Once the processor has been delidded, a bunch of people have been applying their heatsinks directly to the die itself instead of putting the lid back on. I'd be in this boat myself. With liquid metal TIM and a decent waterblock, we might be seeing third and fourth gen chips pushing 1.5V-1.6V.

Atleast on Sandy Bridge, cpu degradation will be accelerated a lot no matter of temperatures above 1.4-5V. Heat is only one part of it when overvolting quite a bit.

2013-07-13, 23:06	#6
kracker "Mr. Meeseeks" Jan 2012 California, USA 37·59 Posts	In general I don't let my cpu go above ~60-70C, over that it makes me really uncomfortable. And not defending AMD at all here, but I have run 24 hours of P95 stress test on a AMD quad at the mid 70's without a problem. EDIT2: My Sandy Bridge(mobile) throttles down when it gets to 100C. Last fiddled with by kracker on 2013-07-13 at 23:08 Reason: forgot space

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
running mprime on 4 cores on 2 assignments?	tha	Software	9	2015-01-23 20:53
Workers, Threads, Helper Threads, Cores, Affinity.	lorgix	Information & Answers	12	2011-01-13 22:31
Problems running ECM on 2 workers	MatWur-S530113	PrimeNet	0	2009-01-06 12:02
Running on 4 Cores	Unregistered	Information & Answers	9	2008-09-25 00:53
Heat problems	michaf	Sierpinski/Riesel Base 5	3	2006-07-04 21:11

2013-07-12, 16:04	#1
MacMagnus Jun 2013 6₈ Posts	Heat problems and running less workers than cores I think I have a little heat problem when running LL on 4/4 cores (the outdoor temperature is high now in the summer, causing less cooling in computer). It is around 90 degrees Celsius. How problemativ is it to run the computer at 90 degrees 24/7? Will it be destroyed anytime soon? I tried running 3 workers (on the 4 cores). The result were much better temperature readings. But, when using the program "CoreTemp", I noticed that the cores shared the workload (that is, averaging about 75% load on each core). I didnt think Prime95 were designed with parallel-programmed algorithms? Does running 3/4 cores mean 25% less work done than 4/4, or is it less efficient some way? (I suppose the workers share the cores running 4/4 also...)

2013-07-12, 16:18	#2
kracker "Mr. Meeseeks" Jan 2012 California, USA 37·59 Posts	What CPU, what cooler? 90C is hot.

2013-07-12, 16:27	#3
firejuggler "Vincent" Apr 2010 Over the rainbow 5550₈ Posts	the reason is maybe because running 4 core @ 75% is colder than 3 @ 100?

2013-07-13, 04:19	#4
Xyzzy Aug 2002 20722₈ Posts	http://www.mersenneforum.org/showthr...446#post269446

2013-07-13, 22:11	#5
TheMawn May 2013 East. Always East. 1727₁₀ Posts	90C is definitely some sort of Intel processor as all AMD's I've heard of get pissed off far above 60C. Most of the Core i processors are set to throttle at 105C but I would not want to go anywhere near that. I don't know how the helper threads work. I only encountered them by accident myself when I was testing how memory bandwidth starved my system was. Running three workers is probably your safest bet. If we knew your processor model and frequency and your memory frequency, we could even tell you if you would expect to see much of a decrease in throughput by running three workers instead of four. Prime95 has very little parallel programming, but what you will some times see is the computer distributing the 3 workers relatively evenly over all available cores. This isn't parallel programming. It's just the program handing out work fairly randomly between processors. This will relieve some of the heat issues too, as you will have all the cores at about the same temperature instead of having three cores ~10C warmer than the other. If 90C is due to abnormally high outdoor temperatures then you're probably alright to keep it running. If it is due to normal temperatures for the season, you'll want to figure out some sort of solution. We've had a couple of very warm days here which brought my CPU from an average of 72C and max of 75C to an average of 80C and a max of 85C, but this was only during maybe 6 hours of the day. This didn't worry me. If it was going to be touching 80C regularly for three or four months, I would be a little bit concerned. Having a k-series processor, I can just clock it down from my current overclocked state. Underclocking yours could help too, but this will also affect the performance of your computer across all your applications. I've heard somewhere that your processor's life expectancy is roughly halved for every 10C. Thermal throttling occurs at 105C so Intel is probably telling you that 100C is "okay", but this is "okay" as far as the warranty is concerned. If your warranty is good for 12 months, then your processor may last 24 months at 90C, 48 months at 80C, etc.. Bear in mind this is a very rough rule of thumb.

2013-07-14, 02:32	#9
kladner "Kieren" Jul 2011 In My Own Galaxy! 2·3·1,693 Posts	I'm not happy if my AMD FX gets close to 60 C. I know Piledrivers are less temperature sensitive than the Thuban I had before. But Core i-x chips are still more tolerant. It seems that Intel might have decided to use up some of that inherent margin by switching to a cheaper, less effective die-to-heat-spreader interface material. It is easy to imagine that adhesive, non-molten-metal compounds are easier to apply acceptably than solder. I would hazard that material cost is less significant than manufacturing cost in this case. It will be interesting to see if the solder interface gets reintroduced in some supreme-ultra-premium-clocked (priced) CPUs. As pointed out earlier, the reduced die size of recent Intel chips already makes the interface a choke point. That said, the "delidding" procedure seems all too risky, with a small misstep yielding a dead chip. And, getting evenly distributed force of the right amount is harder on a smaller, more fragile target.

2013-07-14, 18:24	#10
TheMawn May 2013 East. Always East. 6BF₁₆ Posts	The de-lidding process is something I would be interested in testing out myself at some point. Once the liquid metal TIMs start catching on, a person will see some MAJOR improvements in getting away from the current crap paste --> IHS --> more crap paste --> stock heatsink. Once the processor has been delidded, a bunch of people have been applying their heatsinks directly to the die itself instead of putting the lid back on. I'd be in this boat myself. With liquid metal TIM and a decent waterblock, we might be seeing third and fourth gen chips pushing 1.5V-1.6V. Will still need to wait for quad channel IB-E processors or for DDR4 / GDDR5 memory to take advantage of this though. 4.6GHz i5-3570k is bandwidth limited by 2400MHz memory, so there isn't much point in trying to push 6.0GHz on the same memory.