Sign in with
Sign up | Sign in
Your question

The Truth About PC Power Consumption

Last response: in CPUs
Share
October 19, 2007 5:52:49 PM

AMD and Intel stress idle and max power use by their processors and platforms. To tell the whole story, we look at how much power typical systems consume over time, under different workloads.

http://www.tomshardware.com/2007/10/19/the_truth_about_pc_power_consumption/index.html
October 19, 2007 6:02:45 PM

Its already posted, and unfortunately, flamed.
October 19, 2007 6:59:25 PM

Duh, P = E / t ... so E = P / t for constant P situations, otherwise E = Integral ( P dt), so you can decrease energy used by making the P/t fraction lower...
Related resources
October 19, 2007 10:03:36 PM

What I dont get is why a 570w power supply was used for a system that barely went over 200w at full load. (or barely over 150w in the intel case).

October 19, 2007 11:03:47 PM

kittle said:
What I dont get is why a 570w power supply was used for a system that barely went over 200w at full load. (or barely over 150w in the intel case).


That's easy. Take AMD for example. They recommend a 550 Watt PSU with at least 35 amps for their 2600xt DDR3 series. Any questions?
October 20, 2007 2:03:35 AM

Also, with a smaller amount of power being drawn from a higher wattage power supply, you get a more stable power being delivered ... in addition to the efficiency of the power supply being better (ie more of the power being drawn from the wall is actually used in the system in relation to being turned straight into heat).
a b à CPUs
October 20, 2007 3:51:51 AM

kittle said:
What I dont get is why a 570w power supply was used for a system that barely went over 200w at full load. (or barely over 150w in the intel case).

Other than stable power that meets the systems needs and maximum PSU efficiency, the choice of PSU wattage isn't important for this article.
a c 125 à CPUs
October 20, 2007 3:56:06 AM

Actually a 300-350(provided it meets the needed 12 volts requirments) watt psu would have been the most efficient since power supplies hit there map efficeny at about 50% load

Corsair shows it best


depending on the psu peak efficiency happens in the 45%-55% area

And the power is no more or less stable as long as its a good psu...

I have been trying to hammer this point home forever....a GOOD psu does not need to be 1000 watts to run a modern system

My system takes at most 350 watts from the wall(and thus does not get 24/7 use....i have another system for that)

Case: Antec 900
PSU: OCZ 700watt(got for the efficiency)
Board: P35 DS3R
CPU: Q6600 @ 3.00
CPU cooling : Zalman 9500 @ 500-2200 rpms(PWM)
Memory: 2x 1024MB ,2X 512MB 667 @ 4,4,4,10 2.0 volts
Video: 8800GTX 600(core) 1400(shader) 900 x2(memory)
Storage:
2x WDC 250gig (Raid0)
2x Seagate 320gig(Storage)
2x Seagate 500gig(Storage)
Optical drive: Samsung SH-W162
TV card: PVR 250
Sound: Creative SB Audigy 2zs

That said a 2600XT does not need 35 amps. It takes under 4 amps(vs the 11 used by a 8800GTX) of 12 volt power to run. so unless you are running some super high end 4 socket setup 35 seems to be a high rating...
a c 203 à CPUs
October 20, 2007 4:05:11 AM

Interesting review. Thanks for putting it up.
October 20, 2007 4:57:52 AM

coret said:
Also, with a smaller amount of power being drawn from a higher wattage power supply, you get a more stable power being delivered ... in addition to the efficiency of the power supply being better (ie more of the power being drawn from the wall is actually used in the system in relation to being turned straight into heat).


Coret, your conclusion here is wrong. It is actually more efficient to have a lower total power supply because it tends to have an efficiency curve that is domed. Where the it is most efficient in the middle of its range.

Take for example the new Energy Star requirements. It make a system that meets CAT A idle = 50W you would need to use a PS that is rated at 250W or less. This would be using onboard graphics a single stick of 512MB memory and a processor that idles at or below 8W. The Energy Star requirements demand the use of an 80Plus efficient PS. The definition for this is it must be 80% from 20% loading up to 100% loading. So a Power Supply that is 250W running at 20% load is equal to the system running at 50W.

I have measured these systems in my work. If you move below the 20% loading the efficiency really drops of. So your statement that that about the efficiency of a High wattage Power supply is the opposite of what you have stated. And your other statement that about more stable power is also called into question.

Most Power Supples need a minimum loading for them to regulate accurately. If your high wattage PS is not provided with a sufficient load you could have fluctuating power delivery.

Most PS vendors provide an power to efficiency graph on their websites. If you want the best tuned system then match your power supply's most efficient output to your most typical loading.
a b à CPUs
October 20, 2007 5:26:18 AM

If you build a new computer now, chances are it will have a 500W PSU or stronger, not 300W. I think they chose well, matching real life better.
October 20, 2007 5:29:47 AM

aevm said:
If you build a new computer now, chances are it will have a 500W PSU or stronger, not 300W. I think they chose well, matching real life better.


That is true but to be in the PS sweet spot you would need to run your system at something other than idle. You system would need to draw about 250W. That means a either a 2900XT or a 8800GTX and a quad core processor.

October 20, 2007 5:29:59 AM

This was a good review.

I regret to say that I've been saying this for a VERY long time in quite a few places. I love reading those articles where "Dr. XYZ" says "OMG AMZ R0xx0rs for efficiency". My whole take on it is really nobody can say who is more efficient than anyone else because NOBODY has ever had 2 systems run side by side in real world applications where both have to do the exact same tasks. I want to see someone set up 2 desktops/servers and have them both do the exact same work for a month straight. I'm not sure if this is even possible or feasible, but it would be cool to see. Pit an AMD and and Intel chip against each other competing for 'efficiency'.

On another note, I'm glad to see that someone actually crunched the numbers to see how much it differs. I've always thought the ultra low voltage CPUs was crap because if I wanted to do something complex, it would take me longer and could possibly negate the whole purpose for spending money on an 'efficient' processor.

Now I want to go find all those places where I said exactly what this article said and slam all those people that said 'well.. if it went like that then we'd know' and 'why don't you do it yourself and prove it to the world'.

/victory dance
October 20, 2007 5:34:57 AM

All I can say is, what I said on the other thread. If you want power efficiency, don't get a high end video card. :lol: 

I've done my simple test using the P3 killowatt I got a long while back. To compare my results:

My old system P4 3ghz 6800GS 450w PSU:
Idle - 110-115w
My dads new system E4300 @ 2.4ghz 7300GT 420w PSU
Idle - 107-109w
My new :cry:  (screamer) E4400 @ 2.8ghz 8800GTS 620w PSU
Idle - 160-167w

So my readings are right off the PSU end. I also ran a different test, since I got new batteries for my APC 1400 UPS.

The P4 linux system ran 20 mins until it started beeping.

The E4400 XP system ran 11 mins until it started beeping.

Now I dunno about that 2900XT which has 6pin and 8pin for power, like this picture shows from newegg:



So regardless of how power efficiently a CPU maybe, its not the only part that tends to eat power.

Just my 2 pieces of copper. :D 

Edit:

And I did try to underclock my GPU:

Default speeds - GPU@588 Memory@920


Reduced all the way - GPU@295 Memory@460


Power consumption only reflected 12watt difference changing the 2d clocks.

And is still 35watt more then my old P4. :lol: 
October 20, 2007 5:40:12 AM

Grimmy said:
All I can say is, what I said on the other thread. If you want power efficiency, don't get a high end video card. :lol: 

I've done my simple test using the P3 killowatt I got a long while back. To compare my results:

My old system P4 3ghz 6800GS 450w PSU:
Idle - 110-115w
My dads new system E4300 @ 2.4ghz 7300GT 420w PSU
Idle - 107-109w
My new :cry:  (screamer) E4400 @ 2.8ghz 8800GTS 620w PSU
Idle - 160-167w

So my readings are right off the PSU end. I also ran a different test, since I got new batteries for my APC 1400 UPS.

The P4 linux system ran 20 mins until it started beeping.

The E4400 XP system ran 11 mins until it started beeping.

Now I dunno about that 2900XT which has 6pin and 8pin for power, like this picture shows from newegg:

http://images10.newegg.com/NeweggImage/productimage/14-103-036-04.jpg

So regardless of how power efficiently a CPU maybe, its not the only part that tends to eat power.

Just my 2 pieces of copper. :D 

Edit:

And I did try to underclock my GPU:

Default speeds - GPU@588 Memory@920
http://members.cox.net/fade.2.black/temp/8800-p3-nor.jpg

Reduced all the way - GPU@295 Memory@460
http://members.cox.net/fade.2.black/temp/8800-p3-red.jpg

Power consumption only reflected 12watt difference changing the 2d clocks.

And is still 35watt more then my old P4. :lol: 


Grimmy, what chipset does your motherboard use?
October 20, 2007 5:42:47 AM

Lets see...

Asus P4PE. Chipset, Intel 845PE - my old P4 system
MSI SLI 650i nvidia - Dad's E4300
MSI Platinum 650i nvidia - My E4400

To add info for TDP

P4 3ghz - 89w
E4300 - 65w but oc 2.4ghz = 80w
E4400 - 65w but oc 2.8ghz = 84w

Edit:

speedstep enabled

E4400 - 1.6ghz idle = 50w
E4300 - 1.6ghz idle = 53w

using that PSU calculator overclock thingy. :D 
a c 125 à CPUs
October 20, 2007 5:42:53 AM

I need to actually clip my computer(with a fluke meter that uses a lazy ass induction clip, so its easy....i am just lazy) since it will be more accurate then my UPS....
October 20, 2007 5:55:00 AM

Grimmy said:
Lets see...

Asus P4PE. Chipset, Intel 845PE - my old P4 system
MSI SLI 650i nvidia - Dad's E4300
MSI Platinum 650i nvidia - My E4400

To add info for TDP

P4 3ghz - 89w
E4300 - 65w but oc 2.4ghz = 80w
E4400 - 65w but oc 2.8ghz = 84w

Edit:

speedstep enabled

E4400 - 1.6ghz idle = 50w
E4300 - 1.6ghz idle = 53w

using that PSU calculator overclock thingy. :D 


Well, the Nvidia chips are known to not be the best at using power efficiently. Only the latest Intel chipsets based on the series 3 I would recommend if you are looking to really save on system power. And yes, the 8800 GPU's eat much more power.

I upgraded to a 8800GTX 320 from and old 6800. I have not tested my system for power but the air out of the chassis feels warmer.
October 20, 2007 5:57:47 AM

If you want Grimmy, I can tell you how to make a Intel system that uses less than 45W at the wall when in idle. You won't be playing any tasking games on it but it would work well as a HTPC I would think.
a b à CPUs
October 20, 2007 5:59:03 AM

I liked the article.

Now here's a nasty question:

If system A finishes a job in 9 hours and consumes 3 KWh, while system B finishes the same job in 8 hours and consumes 4 KWh, and both cost the same, which one would you rather buy?

Software is usually badly designed, you can't just start it and go home, you need to sit there and press keys from time to time. The 15 cents you save in electricity with system A will cost you an hour of overtime pay (let's say $30) to be paid to an employee to sit there and press the keys. That is, the whole power efficiency discussion is pointless. Get the fastest computer you can, and your total (electricity + labor) costs will be smaller.
October 20, 2007 6:01:29 AM

Hehe... well, I didn't exactly know nvidia chipsets are power hogs, but it did better then my old trusty P4PE MB. But again that points out that CPU's are not the only part to look at. Oh, forgot to mention, the 420w PSU on my dads E4300 system, that PSU is approx 5-6 years old. :oops: 

Man, I wish I could see the difference on the GTX compared to the GTS.

Ya, my 8800 GTS will keep at least on foot warm in the winter. :lol: 
October 20, 2007 6:17:02 AM

pausert20 said:
If you want Grimmy, I can tell you how to make a Intel system that uses less than 45W at the wall when in idle. You won't be playing any tasking games on it but it would work well as a HTPC I would think.


Naw.. thats okay. I'm just surprized my new system eat 50w more then my old system, because of the video card.

I mean, I knew the 8800 would consume more power when its actually rendering a game, but not at idle? Gah... :heink: 
October 20, 2007 6:29:21 AM

Grimmy said:
Naw.. thats okay. I'm just surprized my new system eat 50w more then my old system, because of the video card.

I mean, I knew the 8800 would consume more power when its actually rendering a game, but not at idle? Gah... :heink: 


I believe that if you use the new ATI chipset with one of the 45W DTP AMD processors I could build a system that idles at around 38 to 39 watts.

You give up a lot of performance there but you would have a really low power system. :D 

The E4300 and E4400 are both rated at 12W idle with the newer E4500 rated at 8W idle. In my testing all of the E4xx processors only used about 4.7w or less at idle.

I have even seen a E2xxx idle around 1.5w. If you get one of those you can really build a low power system but you would need to build with a PS that only does 200W output.
October 21, 2007 8:26:22 PM

I wish they would have gone the extra mile and calculated how much this energy savings would equal in terms of dollars. Again some assumptions could have been made based on the modeling that was performed under the tests, and Im sure a dollar amount for energy cost could have been calculated.
October 22, 2007 2:03:30 PM

aevm said:
I liked the article.

Get the fastest computer you can, and your total (electricity + labor) costs will be smaller.


That's not at all clear. Supercomputers take up room, electricity, A/C, etc. not to mention specialized programming. Total costs will probably not be lower unless you have a task that would otherwise take decades on conventional machines (though I suspect a distributed model would probably still serve you better than the supercomputation route unless you're manipulating large matrices). So it's not black and white.

I think energy efficiency for computers is like mpg for cars - it makes a lot of sense to consider it, along with other factors, when making a purchase.
October 22, 2007 2:56:52 PM

This is a bit confusing, let say I have a 500W PSU and when under load the system consumes about 300W... thus turning the rest of the 200W into heat, right?. In this case, we are paying for 500W of power instead of 300W, correct? 200W = waste???
October 22, 2007 3:03:41 PM

Kinda trying to understand your question directly...

If you have a 500W PSU, it doesn't draw 500W when its turned on. It only supplies what the PC's hardware needs.

At idle, because of the 8800GTS, it uses 160W to 167W (I have adjustable fans on a controller which can change that amount). But under a load, it's peak is around 270W. And I'm using a 620W PSU.
a c 125 à CPUs
October 22, 2007 4:44:02 PM

lotiman said:
This is a bit confusing, let say I have a 500W PSU and when under load the system consumes about 300W... thus turning the rest of the 200W into heat, right?. In this case, we are paying for 500W of power instead of 300W, correct? 200W = waste???


The extra power is no wasted, its not even created(converted and stepped down). In the future if you need it you have it....

It will not turn the test of it into heat, the heat is not taken into consideration on the power out put.....a 75% efficient psu will spit out about 75 watts of heat on a 300 watt load, but the extra 200 is still usable....

All psu's use more AC(plug power) then they make DC. There is always loss in heat. This is normal and psu's are wasting less power then ever before with 80-90% efficiencies(vs the 60-70 of just a few years ago).

So a 80% efficient PSU will waste about 20% of its total power on heat....so the 80% is your DC and the 20% is your heat/energy loss.

Think of a 500 watt load on 2 psu

1. 65% efficient....
2. 85% efficient....

PSU 1 will take 500 watts + 35%(loss) = 675 watts at the plug
PSU 2 will take 500 watts _ 15%(loss) = 575 watts at the plug

So as you can see things are getting better....
October 22, 2007 8:27:02 PM

I do not understand why you continuously and repeatedly put a $120 part up against a $200 part and then pretend like that is insignificant. Why not compare the E4300 with the X2-5000? Hell, the article might have even made more sense if you had done this. (Depending on who you're shilling for I guess.) All I know is that you've discredited yourself for the last time. Dealing with bias is one thing, but when your bias becomes destructive to the actual content, that is when it is time to scratch this site off my list.
October 22, 2007 8:30:51 PM

The article is biased in several ways, the biggest way being that they used the most powerhungry AM2 chipset for the amd system, while they also could have used the amd 690g chipset.
Besides this they could have used an EE AMD chip, and last they could have looked at price like shadowmaster pointed out.
October 23, 2007 4:00:26 AM

Doh! Haven't logged into forums for a long time (I still visit site regularly though)! Anyways, I am very glad someone finally decided to look into total power consumption. I am slightly disappointed as I do think the article is biased because:

Quote:
The delay of new processor products Phenom X2 and X4 forced AMD to shift its focus from "look at how fast our products are" to "look at the value we provide".




We were told AMD is selling a value product and the only way the X2 5000+ consume less power than a E6400 is to outperform it and return back to idle much quicker? And even then the X2 5000+ would have to have a pretty wide margin because its idle usage is higher than the E6400... The X2 5000+ clearly did not have a shot to begin with in my mind.


From my experience, first thing I do is determine my maximum load. What is the absolute slowest computer my most intensive app can run on satisfactorily? You may want to add some room for newer apps.

Second thing I would do is monitor my usage over course of hours, days, and a week. Fortunately for me, I know pretty well my usage from experience.

And finally a whole lot of extrapolation or actual measurements and number crunching. I generally simplify the process by calculating based on time spent at or below 25%, 50%, 75%, and 100% load.

FYI, I am currently using a Turion MT-32 and 90% of the time load is under 25%. :sleep:  On the other hand the Via C3 (Nehemiah) was worst than a P3 because it was constantly under high load...

EDIT: Fix typos/sentences/etc.
October 23, 2007 2:03:45 PM

The only other thought this thread provokes in my mind....

If we ever.... ran desktops on batteries, we all definately wouldn't be spending all this time talking about it. :lol: . o O (brb... need to charge my batteries)
October 25, 2007 3:58:07 PM

I thought the article did a great job of showing the difference in power consumption, both from idle and full load and to the faster CPU finishing 1st and returning to idle. Not only was the E6400 more efficient when running, but it also was faster, thus saving more energy in the long run.

Since AMD hasn't come out with a new desktop processor in ages, I don't think it matters which model is used for comparison. Only the tightest budgets are buying Intel CPUs that compare to AMD prices at this point. Most people would be getting the E6420 or above at this point.

AMD has been making a living off energy efficiency claims since the C2D became available. Nice to see that the C2D was the more efficient and cheaper (in the long run) option all along, and pretty much as I suspected.

Now is it only me that sees the hypocracy of bragging about power efficiency on CPUs and then providing a mid-range video card that uses more power than the competition's high end card? :s
October 25, 2007 4:17:56 PM

Id be really interested to see the power consumption under XP vs Vista. Vista seems to always be doing more in the background. I bet its a far less efficient OS.
October 26, 2007 12:28:29 AM

rallyimprezive said:
Id be really interested to see the power consumption under XP vs Vista. Vista seems to always be doing more in the background. I bet its a far less efficient OS.

I like where this is going!

edit: I am being sarcastic.
October 26, 2007 3:17:08 AM

I don't dispute the results comparing a E6400 and a X2 5000+ however I was disappointed at the article after reading this on the first page:

Quote:
Our take is as follows: power consumption measurements of system idle power, or while a system is busy working on a heavy workload, create a snapshot of minimum/idle power and of the maximum power requirement. As already mentioned, though, this will only help give a feeling for these two extremes. The idle power is the baseline, which people use to start measuring power consumption. Meanwhile, the system or component performance is not considered at all - but it should be, as real-life application scenarios will always create a certain workload. This will always put components into a more energy-demanding state, but also result in quicker completion of a workload when faster hardware is used.
(my emphasis)

How does the test setup in anyway test that?!


(Yeah, I used this in previous reply already. :kaola:  )

Or are you going to claim the X2 5000+ was faster than the E6400? Clearly the E6400 is faster and more efficient than a X2 5000+. :) 


I was hoping they would take a AMD 35/45 TDP EE processor and matched it against an E4300/E6400 with maybe one set benchmarks going very easy on the AMD EE processors and one set like in article that keeps the X2 5000+ on high load.
October 26, 2007 8:51:18 AM

dark41 said:
I thought the article did a great job of showing the difference in power consumption, both from idle and full load and to the faster CPU finishing 1st and returning to idle. Not only was the E6400 more efficient when running, but it also was faster, thus saving more energy in the long run.

Since AMD hasn't come out with a new desktop processor in ages, I don't think it matters which model is used for comparison. Only the tightest budgets are buying Intel CPUs that compare to AMD prices at this point. Most people would be getting the E6420 or above at this point.

AMD has been making a living off energy efficiency claims since the C2D became available. Nice to see that the C2D was the more efficient and cheaper (in the long run) option all along, and pretty much as I suspected.

Now is it only me that sees the hypocracy of bragging about power efficiency on CPUs and then providing a mid-range video card that uses more power than the competition's high end card? :s


Dude seriously....
1. They used the most powerhungry chipset for the AMD system.
2. AMD has loads of more recent and less powerhungry desktop processors.
3. Even a faster, more EE AMD processor is cheaper than the intel processor used.
4. AMD's bragging is about Barcelona

The article plain sucks, because it is biased.
It's like making an article about GPU performance and putting a 3 Ghz core duo in the NVIDIA system, and a 2Ghz P4 HT in the ATI system.
It just makes no sense.
October 26, 2007 9:43:15 AM

Why can't motherboard manufactures with on board graphics make them so when I'm surfing the net/word processing, they use only their on board graphics (cheaper to run plus less wear on my gtx).

Then when the time comes for a spot of gaming it reverts to my PCI-E 8800gtx?

Not that I've got a GTX yet, but you get my drift ;o)
October 26, 2007 9:52:50 AM

The article made it's point, however it can't be used to compare Intel processors with AMD processors in general.

I have some remarks though:
1. The article is about power consumption in general but is used to show that the Intel system used is more energy efficient than the AMD one used and only indirectly proving the ideas behind actual energy savings.
2. A more professional touch would have been better (for example the author should refrain himself from saying: "The reason is Intel's Core 2 processor family beating the daylights out of the aged Athlon 64 family").
3. The audio system used is not an "onboard" solution for the Asus motherboard. It should have been specified in the test setup even though it does not modify the results significantly.

Overall I enjoyed the article. Thanks.
October 26, 2007 1:27:16 PM

Pigbristle said:
Why can't motherboard manufactures with on board graphics make them so when I'm surfing the net/word processing, they use only their on board graphics (cheaper to run plus less wear on my gtx).

Then when the time comes for a spot of gaming it reverts to my PCI-E 8800gtx?

Not that I've got a GTX yet, but you get my drift ;o)


I'd say, cost and design. Onboard video has its own vga port, while videocard these days have vga or dvi ports. Not to mention there are so many manufactures out there, that would have to talk to each other to make things work the way your mentioning.

I remember when I had the voodoo monster card way back. That was when it actually did the 3d graphics in pair with your video card, which I had a ATI card, can't seem to remember it. But it was kinda based on just using a reg card for 2d, and the voodoo for 3d. :lol: 
January 24, 2008 10:44:14 AM

The test is pretty subjective and although a half reasonable attempt to keep the components the same was made, clearly there were other more efficient platform configurations that could have been used. The conclusion could have read, because I have the fastest GPU I can finish a game benchmark in half the time so I have the most efficient system.

In reality this is a guide only that takes into consideration two as identical systems as possible where in reality there are many more platforms for office pc's that are much less power hungry.

Since most of us here on the forums are enthusiasts, I suggest we are more skewed toward performance, and just as with cars, a higher performance machine is the most likely to consume more fuel.

It also makes a huge difference if the task completed 14 minutes faster by one machine should the overall time taken by the faster system be 30 minutes or 2 days and how often this task was performed.

Most systems I see spend about 1 poof-teenth of their time under full load and about 95% idle so efficiency comparisons are pretty pointless unless compared in idle mode in which state most pc's find themselves most of the time. Except for servers of course, but did that yield the wrong results for an AMD bash.
!