The GeForce GTX 480 Update: 3-Way SLI, 3D Vision, And Noise

Power Consumption

Of course, we all know that there’s a lot more to this story than price and performance. Power is a third critical consideration here.

In my initial review, The GeForce GTX 480 and Radeon HD 5870 weren’t too far apart at idle. The delta under load was quite a bit larger—but that was using FurMark.

We know for a fact (because AMD has told us) that the 5000-series cards are protected from extraneous thermal load by two features: a thermal protection mechanism that throttles clock speed beyond a set ASIC temp, and a regulator feedback signal also able to reduce clocks in response to a potentially damaging situation.

Thus, I swapped off of FurMark for this little update and went back to 3DMark’s Perlin noise test. It’s not as demanding, but it does present a fairly consistent, er, bad-case scenario. The result isn’t as extreme as the review; Nvidia’s GeForce GTX 480 uses about 64W more than the Radeon HD 5870 under load.

Adding a second card to the equation increases system power use by 337W (well over the rated board power for a single GTX 480, mind you) at load. Meanwhile, a second Radeon HD 5870 adds 162W. That's a more difficult figure to swallow (in fact, it's hard to explain, since the GTX 480 is supposed to be 250W board).

This thread is closed for comments
144 comments
    Your comment
  • anonymous x
    Why don't you overclock that cpu higher? Only 3.3 Ghz? The 3rd GTX 480 looks like it's being bottlenecked. You can see the scaling is excellent at high resolutions with AA (from 1 to 2 to 3 cards), but at lower resolutions without AA there's no gain.
  • lashton
    and why not check it against the 5890, the 480GTX is nvidia fastest card, put it against ATI Fastest Card
  • cangelini
    There's a good chance that more CPU would def. help at the lower resolutions--one of the reasons I chose 2560 for the comparisons at the end ;-) For one reason or another, wasn't having much luck getting the retail i7-930/Eclipse Plus combo to overclock very well.
  • cangelini
    lashtonand why not check it against the 5890, the 480GTX is nvidia fastest card, put it against ATI Fastest Card


    A pair of 5870s is actually going to be faster. Should I swing a second 5970, though, I do think a pair of 5970s vs. the three GTX 480s would be a good comparison!
  • cruiseoveride
    Crappy ATi drivers.
  • lunyone
    Did I miss something, but there is NO mention of the power consumption of the 3 x 480's??
  • I have no clue where you people are getting this "5890" Statement from... They have the 5870, and then the next step up is the 5970... Is that what you're trying to say?
  • cangelini
    stuk1intI have no clue where you people are getting this "5890" Statement from... They have the 5870, and then the next step up is the 5970... Is that what you're trying to say?


    Bleh, it's late and it has been a long weekend. Edited :)
  • SpadeM
    Quad fire with 5850 would have been nice (thinking back to a builder marathon with quad fire) to see if ati's quad cards made any improovement over last years ones. Also i agree with lunyone, I for one would have been interested to see the numbers on load for the 3 nvidia cards with the 800W gold power supply (and maybe a comment from you cris about what power supply u think is best for the job. Either go lower wattage but a high efficiency psu or higher wattage but lower efficiency)

    Anyways, it was a informative article, looking forward to a full 512 sp card from nvidia and the second revision to the fermi core.
  • JeanLuc
    stuk1intI have no clue where you people are getting this "5890" Statement from... They have the 5870, and then the next step up is the 5970... Is that what you're trying to say?


    The only place where the "5890" exists is in ATI's folder under 'What to do if Fermi is good".

    And yes where is the tri-sli power consumption numbers, there's no mention of it's omittance in the analysis.
  • cangelini
    JeanLucThe only place where the "5890" exists is in ATI's folder under 'What to do if Fermi is good".And yes where is the tri-sli power consumption numbers, there's no mention of it's omittance in the analysis.


    Actually, addressed in a couple of different places ;-) As mentioned in the test setup section, the 3-way SLI numbers crested 900W under load. But because this was above the Enermax unit's max. continuous power rating, this almost certainly isn't a very efficient measurement. Not that it matters--the noise of three cards was so loud that it's simply not a realistic combination using the Eclipse Plus board.
  • rajangel
    Who wants to pay $10 for 1 fps?

    I'm going to stick with my old 9600 SLI config that I spent $100 dollars on a few years back.
  • nebun
    lashtonand why not check it against the 5890, the 480GTX is nvidia fastest card, put it against ATI Fastest Card


    dummy...the 5970 is a dual gpu card, that's why it was not compared. as far as since gpu cards go nvidia is the fastest one out ther.
  • simple_inhibition
    i wonder if nvidia strong armed toms into agreeing not to show tri-sli power consumption in exchange for the third card (of the available "tens of thousands" /endsarcasm)
  • sinsear
    lunyoneDid I miss something, but there is NO mention of the power consumption of the 3 x 480's??

    He wasn't able to upload the results before the power outage caused by the setup.
  • cangelini
    simple_inhibitioni wonder if nvidia strong armed toms into agreeing not to show tri-sli power consumption in exchange for the third card (of the available "tens of thousands" /endsarcasm)


    Once again, it's like, mentioned in the story :-P
  • falchard
    What couldn't get your hands on a 3rd HD5870? Also ATI makes Crossfire bridges long enough. I got 2 with my MSI K9A2 Platinum, and another one with my MSI HD5870.

    I do wonder about how the results are presented. Its obviously given in a method that favors nVidia. The only results chosen to be highlighted are the only results that show AMD in a poor position. I think to present these findings to consumers is rather misleading and more information should be furnished so consumers can make a proper decision on a product.

    Right now the most common resolution is 1680x1050. 1920x1200 is becoming more popular as well. At these resolutions even considering a single GTX480 as having a performance advantage is questionable compared to an HD5870. The HD5870 also scales better in those scenario. Yet the basis of your conclusion of consistent SLI scaling comes from the least favorable scenario for the HD5870 on resolutions that are uncommon. I don't think its a mystery why a lower memory card would perform worse doing a post processing effect at a large resolution. However, there is more to a graphics card then memory and its evident the difference in architecture choices at common resolutions.
  • micr0be
    always great articles from u chris, hoping u might take on the 5970 and 2x5970 soon.
  • barmaley
    So, I'm wondering how much of overclocking potential these GTX 480 cards have. Yep, they are running hot as it is. But, what if you throw liquid cooling on that 2X SLI setup to keep the temps at 35C highest. How far would they go then?
  • h83
    For those who are brave enough to peak at the power consumption of 3 480s running at the same time:...

    http://www.guru3d.com/article/geforce-gtx-480-3way-sli-review/26

    P.S. this isn´t for the ones with a weak heart...
  • Moshu78
    A lot of bottleneck-ing in the article (almost all games @ 1920x1200 and lower)... a 980X should be just the thing for this. This could be a really nice reference article but the bottleneck-ing ruins all the results.
    I mean... not even the 2560x1600 can be validated since it's not 100% sure you were not into CPU-bottlenecks even at this resolution.
    Damn! :(
  • backin5
    1) What about temperatures under load? Were you afraid to scare us off with the results, Chris? Or maybe I just somehow missed them when reading?

    2) I'm surprised Nvidia recommends the SilverStone Raven RV02. The simple idea behind it is great, but sadly tests revealed that the design in practice (or at least SilverStone's implementation of it) doesn't achieve remarkable results temperature-wise.

    3) The results for Metro 2033 at 2560x1600 with all different configurations seem to indicate there might be some problems with both the game and the drivers (for both ATI and Nvidia).
  • anonymous x
    falchardWhat couldn't get your hands on a 3rd HD5870? Also ATI makes Crossfire bridges long enough. I got 2 with my MSI K9A2 Platinum, and another one with my MSI HD5870.I do wonder about how the results are presented. Its obviously given in a method that favors nVidia. The only results chosen to be highlighted are the only results that show AMD in a poor position. I think to present these findings to consumers is rather misleading and more information should be furnished so consumers can make a proper decision on a product.Right now the most common resolution is 1680x1050. 1920x1200 is becoming more popular as well. At these resolutions even considering a single GTX480 as having a performance advantage is questionable compared to an HD5870. The HD5870 also scales better in those scenario. Yet the basis of your conclusion of consistent SLI scaling comes from the least favorable scenario for the HD5870 on resolutions that are uncommon. I don't think its a mystery why a lower memory card would perform worse doing a post processing effect at a large resolution. However, there is more to a graphics card then memory and its evident the difference in architecture choices at common resolutions.

    Using 1680 by 1050 would be stupid, as
    1. The target audience for people who can afford sli/triple sli gtx 480s will mostly likely use 1920x1200 and up, possibly with 3d or triple monitors.
    2. It will be severely bottlenecked as the i7 930 used for this review is only at 3.3 Ghz. (bottlenecked results are not accurate portrayals of a video card's performance).
  • Relayer
    Why does the additional GTX480 draw so much power compared to one? An extra 337W almost doubles the entire system draw with one card. Any explanation for that?