GPU Idle Power

They make noise, too.

Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee

Post Reply
pod03
Posts: 77
Joined: Mon Mar 24, 2003 12:36 pm
Location: UK

GPU Idle Power

Post by pod03 » Tue Oct 07, 2014 2:04 pm

Hi All,

Not really a silent PC question but certainly the first place that I thought to ask.

Different review sites get quite different idle power consumption differences between competing cards - to give a few examples:
http://translate.google.co.uk/translate ... annel%3Dsb (http://ht4u.net/reviews/2014/nvidia_max ... ndex24.php)
http://www.techpowerup.com/reviews/Pali ... am/23.html
http://hexus.net/tech/reviews/graphics/ ... am/?page=9
http://www.kitguru.net/components/graph ... review/20/
http://www.hardwarecanucks.com/forum/ha ... iew-9.html
http://www.techpowerup.com/reviews/ASUS ... OC/23.html
http://www.tomshardware.com/reviews/nvi ... 41-12.html

I recognise that there are differences in the test PCs, test bench system, test methodologies; that there will be measurement error and variability plus that there sample variances between cards of the same model and that different designs even by the same manufacturer will introduce consistent differences in idle power consumption. However, especially comparing the techpowerup Palit 970 and hexus Palit 970 reviews these appear to suggest very wide sample variance. Am I missing something? If not, does anybody have suggestions about how to identify and ultimately buy a 970 with lower than average idle power without buying a number of cards and returning them until I get a good one. Does ASIC quality (http://www.techpowerup.com/forums/threa ... ty.189077/) affect idle power significantly or just 3D load power consumption?

For context I am looking to upgrade a vanilla MSI GTX650 graphics card most likely to a GTX970. My key criteria is a low idle power but with a worthwhile improvement in frame rate of the 650 - I personally do not think that a GTX 750Ti is worth it. It will be installed in a DH77EB running a i3-2120 CPU @ 3.30GHz but I may change the processor at some point. A card any longer than 300mm will be difficult to fit in the case. I think the Strix will be too wide but hope that I can get a MSI Gaming 970 in - both should be fine for length. With the integrated graphics this system idles as low as 15W and with the GTX650 installed as low as 20W - so it appears the difference between a good GTX 970 and a bad one might be a difference of perhaps 20W idle and 25W. I would quite like a semi-passive card, especially as the card is just below my power supply (V450S) but this is not essential as I think nearly all 970s are quite at idle.

CA_Steve
Moderator
Posts: 7650
Joined: Thu Oct 06, 2005 4:36 am
Location: St. Louis, MO

Re: GPU Idle Power

Post by CA_Steve » Tue Oct 07, 2014 2:42 pm

GTX 960 is due out in another week or so. You might want to wait and see if it meets your needs..and then address the idle power/load power issue :D

SometimesWarrior
Patron of SPCR
Posts: 700
Joined: Thu Mar 13, 2003 2:38 pm
Location: California, US
Contact:

Re: GPU Idle Power

Post by SometimesWarrior » Wed Oct 08, 2014 12:07 am

I think you mean to compare TechPowerUp and KitGuru's measurements of the Palit 970? They both directly measured the card, whereas Hexus measured only AC wall power. It is surprising to see 25% difference in both idle and load measurements (KG's 8/139/166 W vs. TPU's 10/182/197 W), especially when the two sites have comparable measurements for other tested cards. Tom's Hardware also directly measured a GTX 970 (Gigabyte's) and said it used 20 W at idle, double what nearly everyone else's cards used, and chalked it up to sample variance.

It's possible that the Maxwell cards have extreme variance among identical models. But it seems more likely that some review sites botched their measurements while scrambling to publish: some reviews mention that Nvidia only supplied them with cards 1-2 days before the official launch. They may have forgotten to switch between stock clock and factory overclock, or the card was still warm from a benchmark, or any number of things. As far as model-to-model variance, that could be explained just by the different fan models and speeds.

pod03
Posts: 77
Joined: Mon Mar 24, 2003 12:36 pm
Location: UK

Re: GPU Idle Power

Post by pod03 » Wed Oct 08, 2014 1:26 pm

Thanks both,

SometimesWarrior - thanks for clarifying - sorry yes I did mean the comparison between
http://www.techpowerup.com/reviews/Pali ... am/23.html and
http://www.kitguru.net/components/graph ... review/20/ but
http://hexus.net/tech/reviews/graphics/ ... am/?page=9 also shows the Palit as having the lowest relative idle to those compared.

It appears common for GTX980s to be measured with lower idle than the 970s as well - this may be related to the first batch of 980s being reference and the 970s not so different VRM but I also suspect that something else is going on - perhaps one factor is that in general the better quality chips tend to become the 980s.

Steve related to the last paragraph, if the GTX960 is the same chip but with less working then given the difference between the 970 and 980 I am not holding my breath for a really low idle for the 960. Rumour has it today that the 960 is delayed because the yield of 970 chips is good enough and demand high enough for 970s that Nvidia have no commercial incentive to launch a cheaper card. I am not betting either way, I will wait and see, probably until Christmas, to see what the situation is then.

I think there is a broader problem in getting reliable data to make this type of decision e.g. compare:
http://www.vortez.net/articles_pages/ms ... iew,9.html
http://www.techpowerup.com/reviews/MSI/ ... ng/24.html and
http://www.hardwarecanucks.com/forum/ha ... msi-8.html

Oh well I guess I will just have to hope I get lucky.

Thanks,

Mark

CA_Steve
Moderator
Posts: 7650
Joined: Thu Oct 06, 2005 4:36 am
Location: St. Louis, MO

Re: GPU Idle Power

Post by CA_Steve » Wed Oct 08, 2014 2:16 pm

In the past, the x60 hasn't been a cut-down of the chip used for the x70 and x80. It's been it's own mid-range die. I haven't seen anything pointing to anything different for this release.

Looking to save 2W at idle by comparing different website reviews is probably a fools errand. Test setups are different, measurement accuracy varies, board circuitry varies, etc..the best you can do is to compare boards at one site to minimize variances.

Same with load power.

Post Reply