Sign in with
Sign up | Sign in

Power-Hungry, Hot, And Big

GeForce GTX 295 Vs. GTX 275 SLI: When Two Are Better Than One
By

Performance aside, there are a few other considerations in comparing one card to two.

The first is physical size. This really only works if you have the platform and available PCI Express slots for it. It might be worth paying a little extra for a GeForce GTX 295 in order to save two expansion slots worth of space. Or, if your motherboard isn’t SLI-certified, taking the single-card route saves you from an upgrade.

Next up is heat. We know from experience that running three GeForce GTX 280s is difficult if you don’t have enough airflow. The cards have virtually zero room between them, and two of three are left trying to suck air through thin gaps between the boards. The only way to make thermal management any more difficult is smashing a pair of GPUs onto a pair of PCBs wedged into a double-slot package.

The GeForce GTX 295 pulls it off though, and at roughly the same core temperatures as the hottest card in a GeForce GTX 275 SLI arrangement. Note that, while the GTX 295 does exhaust much of its heated air out the card’s back-panel, an open top means there will also be a fair amount of recirculation in a closed chassis. In contrast, the GTX 275s blow everything out the back.

You would think that, given their identical core and memory configurations, two GeForce GTX 275s would gobble as much power as a GeForce GTX 295. In fact, that’s not the case at all. Consumption is close at idle, varying by about 20W. But under load, the two GTX 275s use almost 100W more than the taxed GTX 295.

So, as we start looking at numbers, bear in mind that two GeForce GTX 275s will eat up twice as many expansion slots, get warmer, and use more power than a single GeForce GTX 295. In return, you can buy those 275s for about $20 less than a regularly-priced 295 (providing you can find one for sale, that is).

Now let’s factor performance into the equation. Remember that the GPUs on a GeForce GTX 295 run at 576 MHz (with 1,242 MHz shaders), while the memory clock is set to 998 MHz (1,998 MT/s effective). Stock GeForce GTX 275s run at 633 MHz (1,404 MHz shaders) with 1,134 MHz memory (2,268 MT/s). That’s a significant theoretical advantage for the single-GPU cards. Onward!

Display all 51 comments.
This thread is closed for comments
  • 0 Hide
    Anonymous , May 13, 2009 6:12 AM
    been waiting for more on the 295...
  • -9 Hide
    dragonfang18 , May 13, 2009 6:24 AM
    I already ordered 2 295's... $504 each.
  • -7 Hide
    imrul , May 13, 2009 6:43 AM
    "Intel Core i7 920 Extreme (Bloomfield)" (page 3)

    there's no extreme version of i7 920, nor is it bloomfield...
  • 6 Hide
    dragonfang18 , May 13, 2009 6:59 AM
    i7 IS Bloomfield imrul...http://en.wikipedia.org/wiki/Intel_Core_i7
  • 5 Hide
    reasonablevoice , May 13, 2009 7:01 AM
    What the hell is up with the underclocked cards out performing the others in that H.A.W.X.

    Can the author of the article comment with what they think is going on there?
  • 0 Hide
    cangelini , May 13, 2009 7:20 AM
    reasonablevoiceWhat the hell is up with the underclocked cards out performing the others in that H.A.W.X.Can the author of the article comment with what they think is going on there?


    Happened in WiC w/o AA as well. Difficult to say went on there, but the results are repeatable. Probably more important, though, is that when more of an emphasis is put on the graphics subsystem, you see those stock-clocked boards take the lead, as we'd expect.
  • 1 Hide
    avatar_raq , May 13, 2009 7:39 AM
    1.Very good article, unlike some other author's articles in this site, this article is solid (starting from the test system down to the conclusion) and interesting, this is what I always expect from Chris.
    2.As for the strange issue in L4D, HAWX and WIC where the slower 275s beat the faster ones....Odd indeed. Is there any chance the normally clocked cards automatically clocked down to 2D mode or somthing in-game? In other words the GPUs usage dropped due to the CPU bottleneck or whatever, and the cards' driver decided to clockdown to save energy! I've seen nvidia and ati cards do that. The monitoring utility of rivatuner could have revealed such things since it shows real-time clocks..BTW what software did you use to downclock?
    It would be funny to consider downclocking our cards to 'gain' performance!!
    3.I hope the new (single PCB) 295 will drop in larger quantities, perhaps it will be more practical than the current one, and will tip the balnce here in its favor.
    http://www.techpowerup.com/img/09-05-12/13c.jpg
  • 8 Hide
    Anonymous , May 13, 2009 8:51 AM
    Can you put the "online shop" section underneath the "Next" button for the next page...its really annoying and inconvenient to have it positioned within the article as it seems to be.
    Thanks
  • 4 Hide
    rags_20 , May 13, 2009 9:12 AM
    Please benchmark in Very High.
  • 1 Hide
    JeanLuc , May 13, 2009 9:41 AM
    rags_20Please benchmark in Very High.


    I noticed that to, if I owned that kind of hardware I would be playing every game at the highest settings even if it is Crysis.
  • 1 Hide
    salemi , May 13, 2009 10:50 AM
    very good Article, thnx
  • -8 Hide
    stlunatic , May 13, 2009 11:11 AM
    Who the f$%& spends 500$ on a stupid graphics card...

    Good article anyhow ;) 
  • 2 Hide
    stlunatic , May 13, 2009 11:12 AM
    Btw why dont you OC that cpu to 3ghz+?

    Who buys and i7 to use it at stock?
  • 0 Hide
    JeanLuc , May 13, 2009 11:15 AM
    stlunaticBtw why dont you OC that cpu to 3ghz+?Who buys and i7 to use it at stock?


    People who worry about their voiding there warranty, people who buy from HP/Dell etc.
  • 1 Hide
    Anonymous , May 13, 2009 11:48 AM
    People who buy an HP/Dell don't buy GTX 275s in SLI...

    I feel many of these games may have been bottlenecked by that CPU. Would've liked to see these tests with 3.5ghz.
  • 4 Hide
    daeros , May 13, 2009 12:27 PM
    Quote:
    The most striking result here is the drop from 1920x1200 to 2560x1600. The same bug seen in Crysis manifests itself here as well.


    This isn't a bug, nor is it fixable by a driver update. It's called not having enough VRAM to handle all those MASSIVE textures at quadruple their on-screen resolution. The same thing happens when I move from 1920x1200 to 2048x1536 on my 4870. The only solutions are smaller textures or more VRAM. This is why the "professional" cards (FireGL/FireSTREAM and Quadro/Tesla) will often have 2-4 times the framebuffer as the desktop counterparts.
  • 1 Hide
    JeanLuc , May 13, 2009 12:34 PM
    DaerosThis isn't a bug, nor is it fixable by a driver update. It's called not having enough VRAM to handle all those MASSIVE textures at quadruple their on-screen resolution. The same thing happens when I move from 1920x1200 to 2048x1536 on my 4870. The only solutions are smaller textures or more VRAM. This is why the "professional" cards (FireGL/FireSTREAM and Quadro/Tesla) will often have 2-4 times the framebuffer as the desktop counterparts.


    IF the problem is VRAM then why isn't the same result replicated in the GTX 275 benchmarks? Bear in mind SLI setups can only address half the available ram so in this case both cards have the same amount of VRAM.
  • -1 Hide
    Anonymous , May 13, 2009 12:38 PM
    It would have been interesting to see the performance of 2 GTX 295 in SLI. Let's face it, if GTX 275 SLI can outperform a single GTX 295, then the _ONLY_ argument left for a 295 is SLI.
  • 1 Hide
    daeros , May 13, 2009 12:46 PM
    Well, let's see: the 275 at stock clocks went from 31.3 to 7.4, the 275 at 295 speeds went from 28.2 to 8.1, and the 295 went from 28.7 to 7.9. So you tell me how it didn't happen in the 275 benchmarks.

    ps- If you want validation of my remarks, just take a look at all the benchmarks comparing the 4870 512MB to the 4870 1GB. You can see the exact same thing there. Or, if you want to stick with the green, go back to when the 8800GT 256MB came out. At low resolutions it was fine; but crank up either the resolution or the AA/AF, and it chokes.
  • 0 Hide
    daeros , May 13, 2009 12:49 PM
    Minor correction to my first post- I meant 1920x1440, not 1920x1200.
Display more comments