TDP vs. actual power consumption with Prime95 running?

ixfd64 · 2012-08-30, 21:50

How does the power consumption of a CPU at full load (e.g., running Prime95) compare to its TDP in general? I know the TDP is the maximum power the computer is required to disperse, and that it is rarely exceeded under normal conditions. However, I'd expect a computationally intensive program like Prime95 could push the CPU power usage to near the TDP. Does anyone here have any figures for comparison?

kladner · 2012-08-30, 22:12

I don't know if this helps or not. The attached are CPUID Hardware Monitor captures. These numbers are for a Phenom II x6 1090T running at 3.2GHz, its stock speed. P95 (Win7-64) is running 2x P-1. mfaktc is running in 4 instances on a GTX 570 with dedicated CPU cores. CUDALucas is running on a GTX 460.

I think the rated dissipation is 125W. The whole system (box only, no monitor) is drawing ~640W from the line at full load. Of course the majority of that is feeding GPUs.

retina · 2012-08-31, 01:57

TDP is not the same as maximum power draw.

I have not measured it, but I would expect that P95 {w|c}ould easily exceed the TDP in most systems.

When running P95 it is probably a good idea to augment a TDP-rated cooling system capability by at least 50% (figure pulled from thin air so adjust as needed for your particular system).

ixfd64 · 2012-08-31, 02:26

Hmm, I was under the impression that it would be hard to exceed the TDP just by running Prime95. For example, a Tom's Hardware team ran four instances of Prime95 on a QX9650 and measured a power usage of only 74 watts (compared to its 130 W TDP). I wonder how they did it?

I have another (dumb) question. I've noticed that some server chips have similar TDP ratings to that of consumer chips despite having more cores, and I guess a reason for this is that server chips often have a lower clock speed compared to their mainstream counterparts. However, it also seems that many server CPUs (at least the high-end Xeons) have no integrated GPU. Would the absence of an on-die GPU allow more CPU cores in its place, or are the power savings negligible in this case?

retina · 2012-08-31, 02:36

According to this article AMD and Intel define TDP differently.

There are many things that can affect power dissipation levels. Voltage, frequency and workload are the main culprits. Generally power dissipated is proportional to frequency cubed. This is because current levels are proportional to voltage squared and linearly to frequency. But as the frequency increases so does the voltage to compensate. So in summary we have:

Watts = a + V²f

And: V = b + cf

Where a, b and c are constants specific to each chip.

Therefore: Watts is proportional to f³

Dubslow · 2012-08-31, 02:45

Quote:

Originally Posted by retina

According to this article AMD and Intel define TDP differently.

Umm... that's from 2004. It's likely that at least one of them uses a different definition now.

retina · 2012-08-31, 02:58

It is probably also worth mentioning the thermal runaway effect. As a chip gets warmer the transistor leakage increases. So even if all others things are kept constant the power levels can increase simply because the chip is getting hotter. Usually this is not particularly noticeable at lower temps but as the chip starts to get over 80C-90C (approximate values here, the actual knee point depends upon the process used) thermal runaway effects become very important. Because of this, even a small change in cooling effectiveness can have a larger effect on chip temperatures and power consumption.

tl;dr - Keep your chip cool and save power

2012-08-30, 21:50	#1
ixfd64 Bemusing Prompter "Danny" Dec 2002 California 2,503 Posts	TDP vs. actual power consumption with Prime95 running? How does the power consumption of a CPU at full load (e.g., running Prime95) compare to its TDP in general? I know the TDP is the maximum power the computer is required to disperse, and that it is rarely exceeded under normal conditions. However, I'd expect a computationally intensive program like Prime95 could push the CPU power usage to near the TDP. Does anyone here have any figures for comparison? Last fiddled with by ixfd64 on 2018-05-22 at 22:01

2012-08-30, 22:12	#2
kladner "Kieren" Jul 2011 In My Own Galaxy! 2·3·1,693 Posts	I don't know if this helps or not. The attached are CPUID Hardware Monitor captures. These numbers are for a Phenom II x6 1090T running at 3.2GHz, its stock speed. P95 (Win7-64) is running 2x P-1. mfaktc is running in 4 instances on a GTX 570 with dedicated CPU cores. CUDALucas is running on a GTX 460. I think the rated dissipation is 125W. The whole system (box only, no monitor) is drawing ~640W from the line at full load. Of course the majority of that is feeding GPUs. Attached Thumbnails

2012-08-31, 02:58	#7
retina Undefined "The unspeakable one" Jun 2006 My evil lair 1A73₁₆ Posts	It is probably also worth mentioning the thermal runaway effect. As a chip gets warmer the transistor leakage increases. So even if all others things are kept constant the power levels can increase simply because the chip is getting hotter. Usually this is not particularly noticeable at lower temps but as the chip starts to get over 80C-90C (approximate values here, the actual knee point depends upon the process used) thermal runaway effects become very important. Because of this, even a small change in cooling effectiveness can have a larger effect on chip temperatures and power consumption. tl;dr - Keep your chip cool and save power Last fiddled with by retina on 2012-08-31 at 03:00

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
power consumption	esakertt	Hardware	1	2012-10-09 20:19
power consumption	Unregistered	Information & Answers	2	2011-01-27 20:48
POWER CONSUMPTION idle versus Prime95	Peter Nelson	Hardware	10	2005-01-16 19:42
Now power consumption numbers using Prime95	Dresdenboy	Hardware	1	2004-11-23 18:29
Power consumption	optim	Hardware	8	2003-12-06 04:13

2012-08-31, 01:57	#3
retina Undefined "The unspeakable one" Jun 2006 My evil lair 3·37·61 Posts	TDP is not the same as maximum power draw. I have not measured it, but I would expect that P95 {w\|c}ould easily exceed the TDP in most systems. When running P95 it is probably a good idea to augment a TDP-rated cooling system capability by at least 50% (figure pulled from thin air so adjust as needed for your particular system).

2012-08-31, 02:26	#4
ixfd64 Bemusing Prompter "Danny" Dec 2002 California 2,503 Posts	Hmm, I was under the impression that it would be hard to exceed the TDP just by running Prime95. For example, a Tom's Hardware team ran four instances of Prime95 on a QX9650 and measured a power usage of only 74 watts (compared to its 130 W TDP). I wonder how they did it? I have another (dumb) question. I've noticed that some server chips have similar TDP ratings to that of consumer chips despite having more cores, and I guess a reason for this is that server chips often have a lower clock speed compared to their mainstream counterparts. However, it also seems that many server CPUs (at least the high-end Xeons) have no integrated GPU. Would the absence of an on-die GPU allow more CPU cores in its place, or are the power savings negligible in this case?

2012-08-31, 02:36	#5
retina Undefined "The unspeakable one" Jun 2006 My evil lair 15163₈ Posts	According to this article AMD and Intel define TDP differently. There are many things that can affect power dissipation levels. Voltage, frequency and workload are the main culprits. Generally power dissipated is proportional to frequency cubed. This is because current levels are proportional to voltage squared and linearly to frequency. But as the frequency increases so does the voltage to compensate. So in summary we have: Watts = a + V²f And: V = b + cf Where a, b and c are constants specific to each chip. Therefore: Watts is proportional to f³