Sign in with
Sign up | Sign in

3D Temperature, Noise, Power Consumption

Comparison: Factory-Overclocked Versus Reference Graphics
By

There is something we’d like to clarify before going into details of 3D operation. If a powerful graphics card holds the GPU temperature at, say, 60 or 80°C, this is almost never the credit of the cooling solution or the graphics card vendor, but the temperature threshold programmed into the cooling profile. Typically, the fan speeding up will create sufficient air flow to tackle GPU heat, and even reference cards are capable of maintaining 60 or 65°C. It's all a matter of fan speed and operating noise. However, each cooler does have its limits.

The programmed GPU temperature threshold of the overclocked card typically is the same or at least similar to the default, which is 80°C in the case of AMD's Radeon HD 5870. In the case of the MSI R5870 Lightning, the threshold is 65°C, and the cooling solution will try to stay below this temperature even if the card is overclocked. Although the temperature threshold should not be seen as a quality item, it is indeed reasonable to drop the temperature threshold if the cooling solution allows, and as long as you don’t increase cooling noise too much.

On the overclocked reference card, the default cooler has to handle the additional heat dissipated by the +50 MHz GPU clock speed. The overclocking has a more noticeable impact on the noise level at +0.5 dB(A) and the temperature increases from 80 to 82°C. The difference is smaller on the factory-overclocked MSI board, which is because of the smaller overclocking margin. The card already runs at 900 MHz, and the additional 25 MHz we could achieve without voltage tweaks are less significant. The cooler with its two fans can easily handle the additional heat caused by our overclocking. Effectively, there is only a little noise difference between the reference card and the optimized overclocking models. The latter offers higher default clock speeds and cools more efficiently, resulting in similar cooling noise as on the reference board.

The power consumption differences between the default clock speeds and the overclocked settings aren’t as significant when looking at peak load conditions. We measured a difference of 10%, and this it typically because of adjusted voltage settings on overclocked graphics cards. This is oftentimes necessary to ensure stable operation.

The last two diagrams show the events during the 13 minute benchmark run. You can see that the graphics processor reaches peak temperature within minutes. Both cooling solutions provide similar results when the benchmark takes a quick break and doesn’t cause GPU load, as the temperature decreases quickly.

The percent value in this diagram has only limited meaning, as it is calculated from the maximum fan speed. The noise level is more relevant. MSI’s Lightning card is a bit better than the reference cooler from AMD. We found the steadiness of the results more impressive, as MSI’s twin-fan cooler is capable of handling sudden temperature differences much quicker than the reference cooler design.

However, you can avoid high noise level or frequent fan speed changes through the Catalyst driver or an overclocking tool, such as MSI’s Afterburner, by simply setting the fan to a fixed speed. Keep in mind that you should look at the GPU temperature if you decide to go this way, as the cooling performance is of course limited by a fixed, low fan speed. There are dynamic fan speed profiles, but we found that this option is sometimes locked, which means that you can typically only choose between the factory auto setting or a fixed fan speed.

Display all 27 comments.
This thread is closed for comments
  • 5 Hide
    chrissmartin , July 25, 2010 11:30 AM
    i will buy a normal card with good cooling and overclock it.
  • 3 Hide
    rohitbaran , July 25, 2010 11:59 AM
    Too less performance gain (4-5 FPS) for the too much extra price paid ($50-$70) along with too much increase in power consumption (30-40 W. Better to buy a reference card unless you want the best no matter what the price is.
  • 1 Hide
    Mark Heath , July 25, 2010 12:53 PM
    I like the idea behind the article and I'm glad that there's been some testing done on it, but I don't really like the article itself.

    There are references to how much the reference card is overclocked by, but there is no clear section about it, you started off with some of the summary charts, instead of putting them after the initial tests and you kept switching what cards you were comparing, with less warning than I would like..

    Plus there's that 2GB 5850. Is that really a 2GB 5850, or is it actually 2GB 5830, because it kept getting beaten by the 1GB 5850..

  • 1 Hide
    Onus , July 25, 2010 1:27 PM
    When I buy a new GPU, I look for one who's reference clocks make it "good enough" as-is; but I do look for a better / quieter cooler.
  • 0 Hide
    mrhoshos96 , July 25, 2010 2:01 PM
    great article but u should have compared the reference against the sapphire vapor-x
  • 8 Hide
    Tesla1483 , July 25, 2010 2:31 PM
    Quote:
    In order to their products apart, third-party vendors take reference GPUs...


    I'm guessing they meant "In order to SET their products apart..."
    Seriously, it's the first sentence in the article. Does no one proof read anything here? Isn't that like the first thing you learn when you want to become a writer? Sorry, rant over.
  • 0 Hide
    youssef 2010 , July 25, 2010 9:23 PM
    "This cooling concept works pretty well, as it not only provides effective graphics component cooling, but it also helps to remove warm air from the inside of your PC. "

    The reference 5870 has a vent on its side so, it doesn't completely remove the hot air from the case but recirculates a part of the hot air back into the case
  • 1 Hide
    duk3 , July 26, 2010 3:01 AM
    It's a 5850, look how close the framerates are.
    The amount of ram doesn't really matter at these resolutions.
  • 4 Hide
    LLJones , July 26, 2010 4:28 AM
    The cards just do not offer enough gain. If you can afford the overclock edition at a premium, then save up a little more and get the next card up or buy two of the next lower cards. The $1000+ dual 5870 from Asus shows what a complete waste of money some, not all, super OC cards are.

    Try this, put together a really really really premium card component wise, and then don't put any cooling on it. Let me decide what I am going to do for cooling and companies can save on packaging, shipping etc due to no cooler.

    If your really aiming at the OC crowd and not just playing a game of marketing, then you know the first thing someone does is to rip off your factory cooler to replace it or at the very least put quality thermal compound on it.
  • 0 Hide
    dEAne , July 26, 2010 6:34 AM
    I like this article, thanks tom.
  • 0 Hide
    bildo123 , July 26, 2010 4:25 PM
    jtt283When I buy a new GPU, I look for one who's reference clocks make it "good enough" as-is; but I do look for a better / quieter cooler.


    Best suggestion, wait for the manufacture to slap on some after market cooling and pay the extra for something useful. It wasn't like this in the past, but now-a-days they like putting on HSF setups that do a relatively poor job.
  • 0 Hide
    kirillian , July 26, 2010 5:11 PM
    You may not have noticed silicondoc, but this past year, NVIDIA really hasn't been bringing its A-game to the table, mostly trying to use marketing to make up for its failures...right now, Red really is performing better in a price/performance setting. There's really little that any honest individual can do to deny that, fan or Red OR Green...it's not Tom's Hardware that's falling down, it's NVIDIA at the moment. Maybe the next gen will be better, but no one knows yet.
  • 0 Hide
    kirillian , July 26, 2010 5:17 PM
    kirillianThere's really little that any honest individual can do to deny that, fan or Red OR Green...


    Edit: ...fan of Red OR Green...
  • 0 Hide
    Adroid , July 26, 2010 6:37 PM
    I have started buying cards from vendors that warranty overclock like EGVA. I paid 159$ for a GTX 260 216 core after rebate well over a year ago, then when I am gaming I use the EVGA precision tool to overclock it to the higher priced cards core / shader speed.

    Its funny because mine will do the exact speed as the factory "superclocked" card's settings at the press of a button - all under lifetime warranty.

    My card before this one was a Radeon, I'm a total unloyal customer. I'm sure they have some similar deals, but EVGA has definetely earned my business when it comes to Nvidia stuff...
  • 0 Hide
    firedust , July 26, 2010 9:01 PM
    I definitely would have liked to see the 460 on here. So far the reviews have been very good.
  • 1 Hide
    itpro , July 26, 2010 11:16 PM
    Wow, you nVidia uberfans really get your panties in a bunch over nothing! The premise of this article was valid: Comparing factory over clocked cards against standard cards. Yes, they chose to compare a mature product in the 5870 rather than a brand new to market nVidia, but the premise was still valid.

    Yes, some of us DO prefer ATI cards over nVidia. So what? For the last two years nVidia has done little more than offer up old cards with new model numbers on them trying to convince us that they were still a competitor, while ATI pushed the envelope in performance, price, and power efficiency. Even now, when nVidia has finally gotten their new chipset out the door and is finally offering a competitive performing card they are still too expensive for what you get and use enough juice to heat a small home.

    If nVidia spins your wheels, then that is fine with me. Personally, I have had more than enough of their crappy drivers and overheating cards to last me a life time.
  • 2 Hide
    d-block , July 27, 2010 12:18 AM
    What kind of moron puts a 5870 in a 32 bit machine? That test pc setup is complete garbage.
  • 2 Hide
    Sykar , July 27, 2010 12:42 AM
    d-blockWhat kind of moron puts a 5870 in a 32 bit machine? That test pc setup is complete garbage.

    more to the point, who the hell in their right mind buys/uses a 32-bit OS anyways?
  • 0 Hide
    randomizer , July 27, 2010 12:43 AM
    Quote:
    What kind of moron puts a 5870 in a 32 bit machine? That test pc setup is complete garbage.

    What kind of person thinks it would make a difference?
  • -1 Hide
    Poisoner , July 27, 2010 4:52 PM
    silicondockirillian, you sad, sad person. Every website in the world has declared the GTX460 the current best bang for the buck PERIOD. You talk about honesty, yet you post the most blatantly dishonest crap of the day one from the red team could cook up. What a sick joke. Just forget it, live in your insane fantasy world of lies and be happy. My God.



    Wow, you must be off your meds. Tom's just posted an article about the awesome SLI scaling of the 460, although it was a week late. I also believe that the lastest graphics card chart came before the GTX 460, although the other GF400s are on the chart.

    As far as the 480 and the 470, I don't think they use that much more power than previous generation cards, or create that much more heat either. I'd say that EVERY review site under rates the higher end Fermi cards. No one seems to get that besides gaming, they are absolute folding MONSTERS. If I folded, I wouldn't use anything less than a 470. To bad nVidia doesn't have an AMD chipset that was in my price range when I was buying a motherboard.
Display more comments