GPU's Energy consumption Energy consumption

I wanted to know how was the measurement of energy consumption of GPU.
Some tips.
Thanks for all,

Bruno C.

I you look up the specs page for a particular GPU, they generally list its maximum power consumption. Actual power consumption in practice will of course depend on what you do with it.

Peter

If you want to know how much extra power the computer uses because of the GPU, I would suggest finding a power meter (the Kill-A-Watt is a great option) and plugging the computer into it. Record the power usage when the computer is on and the CPU is in use, without the GPU installed. Then install the GPU and run a CUDA application. The difference will tell you how much extra power the GPU is causing the computer to consume.

Of course, this is not going to be the same as the power draw measured at the GPU itself, because power supplies are not 100% efficient.

you can also look at 3rd party review sites, e.g. http://www.techreport.com/articles.x/16681/11

So what about the power writting in the box of GPU card?

(for instance, my geforce 8800GT needs 650W, is it maximum power must supply for this card?)

by the way, Hi gatoatigrado.

I admire your Intenet searching capability.

You usually get correctly link.

:)

funny… :) http://forums.nvidia.com/index.php?showtop…mp;#entry530272

I’m not sure I understand your first question. As far as power supplies goes, it’s very important to look at the efficiency of the unit, as well as quality. A 650w power supply, if poor quality, will only deliver 60% efficiency, for effectively 390w. A high quality unit will be around 80% efficiency (90% ones are coming, but they’re probably expensive), which would be 520w, a lot more. Also be warned that some manufacturers test power supplies at unrealistically low temperatures.

“80 plus” certification is good, but sometimes you’ll find supplies without it listed (e.g. the first Corsair power supplies iirc) or which perform similarly. Review sites are always useful. Brand names help if you’re in a hurry: antec (earthwatts only), corsair, seasonic, enermax, thermaltake might be okay, silverstone are all okay (but look at individual unit reviews when you get closer).

80 plus power supplies can be a huge premium over others, but that margin is falling, and you can use techbargains or hotdealsclub to narrow the gap even more.

Nvidia reports more than actual consumtions. Take GTX295, Nvidia figure is 289 watts per card, I have 10 card system, consuming 2,8 KW, including system boards and CPU-s.

More: Estonia Donates GPU supercomputer

Where can I find the specs page for various GPUs? The reason I ask is I am planning to add a GTX285 to my system,

but it needs to be powered by an auxiliary power supply. Is a 300W auxiliary power supply enough for the GTX285?

Yes. The NVIDIA page lists the maximum power usage as 204W:

http://www.nvidia.com/object/product_geforce_gtx_285_us.html

(Click on “Specifications”.)

I found nvidia’s page above the most helpful

No wonder I couldn’t find the specs: The ‘Specifications’ link on the above page did not appear on my Mozilla Firefox 3.0b5 browser. However, when I positioned the mouse in the ‘empty’ space just above the ‘Reviews’ link and clicked, I was able to bring up the spec. page. (I hope I don’t need a fancy browser with the latest Flash player and Java, etc., in order to view Nvidia’s product pages!)

Thanks.

Hi,

This is an amazing beast !!! :)

Did you check whether the GPUs scale perfectly on the same board? i.e. does 4 GTX295 on one board runs 4 times faster

then only one?

thanks

eyal

As in my prev. post, a 300W rated power supply might not deliver anywhere near 300W. My answer would be no, unless it’s 80 plus certified (these power supplies typically have cleaner output as well).

regards,

Nicholas

Good to know that. I’ll try the 300W one first. If it doesn’t work, I can always buy a cheap 400W one.

I highly recommend you instead read the spec’s for the power supply. Undervolting your card could damage it irreparably.

regards,

Nicholas

This is folding machine. Modelling proteins for Stanford University. This task is very parallel and scalable, in my case each GTX295 adds power in linear manner.

Jaak