Sign in with
Sign up | Sign in

Hardware: Core i7 920, Asus X58, Corsair DDR3

Overclocking Core i7: Power Versus Performance
By

It may not be a mainstream system yet, but we decided to use a Core i7 machine for several reasons. First of all, Core i7 is currently the fastest processor available, and we would expect performance enthusiasts to go for it if their budget allows. AMD’s Phenom II would be an almost-equivalent choice, and we've already performed this little experiment on that platform.

The second reason for using Core i7 is its power efficiency. An X58 platform with Core i7 is not a low-power solution at all. In fact, our test system required at least 109 W idle power, which is more than any Core 2 Quad system based on a 45 nm processor and a P45 platform would require. However, Core i7 is far superior to Core 2 when it comes to performance per watt, as the performance benefits usually are larger than the incremental power required.

Our platform of choice is Asus' P6T, which is an enthusiast X58 chipset motherboard. It’s not the company's top-of-the-line model, but it was designed for overclockers and other performance-seeking users nonetheless. We had no problems running Core i7 samples at more than 4.0 GHz using a Core i7 Extreme 965. We used this to ensure that the motherboard would not be bottlenecking higher overclocks on the Core i7 920, which is a much more affordable processor.

We used top tier memory, a very respectable hard drive, and a modern graphics card on our overclocking rig. Our memory was Corsair’s DDR3-1600 Dominator series, the TR3X6G1600C8D, in a 6 GB triple-channel configuration. We did not take full advantage of the possible DDR3-1600 speed, as we started off at the specified DDR3-1066 speed for the regular Core i7 920 (stepping up to the 965 Extreme increases that default to DDR3-1333). System to memory ratio was 2:8 at the 133 MHz base speed, which results in a 533 MHz DDR3-1066 memory speed. We changed the memory multiplier to 2:6 to keep memory speeds within the same levels during our race to 4.0 GHz core clock. Of course, it is possible to fire up the memory voltage and clock speed as well, but this won’t deliver performance gains anywhere near to what a 333 MHz processor clock speed bump can deliver.

Finally, we selected a Zotac GeForce GTX 260² 55 nm graphics card based on the 216-core GPU, which delivers powerful DirectX 10 graphics performance at reasonable power consumption levels. A higher-performance graphics card would deliver better game performance, but it would have decreased the relative power consumption difference in our tests. The Zotac card comes with 896 MB GDDR3 memory, is based on the Nvidia reference design, and comes bundled with Grid.

Ask a Category Expert

Create a new thread in the Reviews comments forum about this subject

Example: Notebook, Android, SSD hard drive

Display all 64 comments.
This thread is closed for comments
  • 6 Hide
    eximious , April 13, 2009 8:14 AM
    Interesting and insightful article. It might be interesting to do a similar analysis when the new D0 stepping is widely available. Thanks for including a variety of video encoding / editing benchmarks too.
  • 0 Hide
    zedx , April 13, 2009 8:14 AM
    No undervolting / underclocking? I'm sure you can greatly improve efficiency by doing something like a clock of around 2.4 ghz and a voltage of around 0.8 - 0.9 volts... Even in default you might be able to undervolt quite a bit...
  • -2 Hide
    tacoslave , April 13, 2009 8:15 AM
    i wish i had one
  • -6 Hide
    onerec , April 13, 2009 9:01 AM
    does this mean that AMD phenom2 is better than Intel when it comes to overclocking?
  • 7 Hide
    falchard , April 13, 2009 10:22 AM
    lol no, this one went from 2.66 to 4.0. In the other comparison with the Phenom II it went from 3.0 to 3.8.
    What I find interesting is that both articles found the magic number to be 3.6 ghz.
  • 5 Hide
    jonpaul37 , April 13, 2009 12:43 PM
    3.6 - 3.8 seems to be the universal sweet-spot for any CPU that allows for the OC. I've OC'd my E8400 to 3.6 and get great results, anything higher might yield an FPS or 2 more, but at the possible expense of more power.
  • 1 Hide
    Anonymous , April 13, 2009 12:46 PM
    Ever since the stock P4 3.6ghz came out, the magic number has been 3.6, every new generation of CPUs since then, people have expected to get somewhat further, but IMO it's safe to say that without some radically new technology, the magic number will always be about 3.6.
  • -4 Hide
    Anonymous , April 13, 2009 12:59 PM
    I will stick with my q9650 overclocked to 4.0Ghz at 1.24v ntm the 8gb ram at 1066 mhz. lolz
  • 4 Hide
    TripGun , April 13, 2009 1:39 PM
    Very good article. However , I do have a problem with the test being ran with an "engineering sample" chip. A lot of boxed I7's won't hit 3.66ghz without at least 1.35 volts. Good read though.
  • 0 Hide
    Anonymous , April 13, 2009 1:40 PM
    Nice article.

    Thanks.
  • 1 Hide
    Pei-chen , April 13, 2009 2:00 PM
    I always believe the sweet spot is half way of stock speed and max OC. In this case, 3.33GHz is the sweet spot while 3.66 means you got a good overclocker. Good read.
  • -2 Hide
    Anonymous , April 13, 2009 2:39 PM
    I second the need for a underclock/undervolt analysis. Also, how about seeing exactly how voltage affects the processor without changing the clock speed. Many people hit 3.8-4Ghz at lower voltages so a base comparison on how much voltage affects the power (theoretically squared relationship) would be helpful...
  • 0 Hide
    Anonymous , April 13, 2009 2:47 PM
    I agree from the results that the safest overclock is 3,33Ghz with turbo mode enabled, for the casual gamer.
    Besides, you will not notice much difference in games running them at 3,33 or 3,66.
    Also the life expectancy of overclocking the CPU @3,33, is longer than @3,66Ghz.

    I wished sometimes 'underclocking' would be done at toms, to see how much power one can save when he's only writing documents, or browsing the web with such a powerful machine. I mean, unless you're a gamer, the computer stays most of the time in a passive mode (if not turned off).

    About the 'magical number',this probably changes with the die. When a processor is created @ 45nm the best results might be 3,6Ghz; but these results should be different on larger or smaller dies (eg: 65nm, 95nm, or the upcoming 40nm, or 32nm). It also has to do with the materials used. Current 45nM processes by intel are done with high metal K gates and stuff, they allow greater overclocking to the standard silicon processors...
  • 0 Hide
    Anonymous , April 13, 2009 2:51 PM
    If everything goes like now, and processors on a smaller die (32 nm, 28nm,...) will be identical copies of current processors, we might see a trend that the smaller the die, the smaller the possible overclocking.
    It all depends on how thick of a layer of insulation Intel uses between the transistors (in on-off switching by lack of words), and if they will invent or discover newer more efficient materials to develop processors or not.
  • 7 Hide
    theJ , April 13, 2009 2:59 PM
    I just ran some calculations on these power consumptions.

    Assuming you keep your computer on 365/24/7, at peak power all the time:
    -$104.49/year for a 2.66 Ghz
    -$204.56/year for 4.0 Ghz

    For a medium usage user: 365 days per year, 4 hours at peak, 20 hours at idle:
    -$61.97/year for 2.66 Ghz
    -$94.60/year for 4.0 Ghz

    For a more modest user: 365 days per year, 2 hours at peak per day, 4 hours at idle:
    -$17.62/year for 2.66 Ghz
    -$29.15/year for 4.0 Ghz

    This assumes 5.6 cents/kWh.

    This is just to give everyone a more convenient way to track power. I know i don't have a feel for 100 W compared to 200 W.
  • 0 Hide
    cadder , April 13, 2009 3:06 PM
    Stock cooling??? I'm not believing that.

    WRT undervolting, maybe we need new mobo features to allow custom "speedstep" features. Run the processor at max. OC speed when it needs it, then drop to stock speed or below when at idle or at low use.
  • 0 Hide
    optimus290 , April 13, 2009 5:05 PM
    these are good performance gains :) . but its not something ground breaking. :|
  • 0 Hide
    mcnuggetofdeath , April 13, 2009 5:07 PM
    cadderStock cooling??? I'm not believing that.WRT undervolting, maybe we need new mobo features to allow custom "speedstep" features. Run the processor at max. OC speed when it needs it, then drop to stock speed or below when at idle or at low use.

    Thats what got me too, with the abundance of good and cheap air cooling solutions negating the effects of Intel's Overspeed protection shouldnt be hard even w/o a board with a BIOS option to that effect.
  • 0 Hide
    funkjunky , April 13, 2009 5:40 PM
    I know I would rather have a bios option for better underclocking, so it is more transparent. Like super speedstep, where it drops the multiplier even more ;) .

    I can't see a reason for tom's to give underclocking numbers, until their are better native dynamic underclocking features on mobos. No one is going to manually under clock their computer when they go and change to writing documents... it needs to be automatic.

    Assuming Thej's numbers are even close to accurate, then having that extra bit of under clocking won't save you anymore than $5-$10, because most core components will still be running taking juice.

    This article was a convenient read, and even more convenient thanks to Thej =).
  • -5 Hide
    funkjunky , April 13, 2009 5:40 PM
    I know I would rather have a bios option for better underclocking, so it is more transparent. Like super speedstep, where it drops the multiplier even more ;) .

    I can't see a reason for tom's to give underclocking numbers, until their are better native dynamic underclocking features on mobos. No one is going to manually under clock their computer when they go and change to writing documents... it needs to be automatic.

    Assuming Thej's numbers are even close to accurate, then having that extra bit of under clocking won't save you anymore than $5-$10, because most core components will still be running taking juice.

    This article was a convenient read, and even more convenient thanks to Thej =).
Display more comments