So how does it feel to have the worst Core i7 980X overclocking scores online?I can't say that it's what I'm most proud of. That award would have to go to the time I wrote a review and left out the 'l' in overclocking. Needless to say after everyone pointed out how abysmal our overclocks were and after even Intel called to let us know that we should be seeing better, we gave it another shot.


Intel's DX58SO Motherboard

A motherboard swap, a little more voltage and backing off the un-core clock a bit yielded something a bit more respectable out of our Core i7 980X sample:

Keep an eye on what we're talking about here. Six cores, 12MB of L3 cache, all running at 4.13GHz with Intel's stock heatsink. With more voltage, even higher frequencies should be possible - but at the expense of increased power consumption.

The performance at 4.13GHz is even more ridiculous than the stock Core i7 980X:

Another ~17% improvement over the already bonkers 980X is just crazy. At 53.7 fps in the second pass of our encoding test we're more than twice as fast as a Core i7 920. As much as we hate that Intel is maintaining two different sockets for its desktop CPUs, the Core i7 980X makes LGA-1366 worth it. Now if Intel could only get that price down.

The performance doesn't come for free though. At 1.359V the 980X draws quite a bit more power:

The 17% performance improvement comes at the expense of a 20% increase in total system power. It's not the most efficient way to get more speed, but if for some reason you're not happy with your 980X's default performance this is the sacrifice you'll have to make.

Thanks for bearing with us as we tried to push our chip further and got some more respectable results :)

Comments Locked

57 Comments

View All Comments

  • masouth - Saturday, March 13, 2010 - link

    Actually that's not entirely true either. Just being plugged in a pc (turned off) and monitor (in standby) are probably drawing anywhere from 7-35 watts depending on the pc, pc power supply switch and the type of monitor.

    The only TRUE way to prevent power draw is to completely unplug everything.
  • jonup - Friday, March 12, 2010 - link

    That's why I hate going back to EU. You worry about stuff like that. You have a government that tells what's right and what's wrong. You need to start spending more and everything will get cheaper so you dont have to worry about 50 watts. And you if you think the environment, even the most pisimistic treehugging scientific forcast give us about 300,000 years before we destroy the earth. That's like 50 times more years that the humans as we know them have been around.

    Back on topic, AT, can you test/overclock with a good aftermarket cooler. First we can see the actual potential of the CPU on air and second we can see how effective is the new Intel cooler. Thanks!
  • busta - Friday, March 12, 2010 - link

    You really need to work on your understanding of science and history

    OR

    stop commenting on things you obviously don't understand.
  • medi01 - Friday, March 12, 2010 - link

    1 ancient
  • vol7ron - Thursday, March 11, 2010 - link

    This is why I keep checking Anandtech every 5 mins :) Screw RSS notifications, manual is better, and AT gets me coming back.

    Perhaps someone has time for some detailed overclocking analysis. That is, not just overclocking the proc, but the combined synergy and sweet spot for overclocking the proc/ram/gpu in such a way that yields the best results.

    By RAM this includes gpu ram and by overclocking, this includes memory timings.

    If only I had the resources, I'd sit down for a week and test, test, test :)


    vol7ron
  • DanNeely - Friday, March 12, 2010 - link

    Tweaking Ram timings, etc in modern CPU architectures doesn't yield anything outside of benchmarks. It was different back in the p4 era where the memory bus was a bottleneck, but since amd64/core architectures that hasn't been the case outside of pathological circumstances. At stock speeds the dual channel LGA1156 controller provides 75% of the worst case memory bandwidth consumption for a quadcore system; a level that only synthetic benchmarks can exceed. For quads LGA1366 provides 100%; for the 6core model it again drops down to about 75%. High end memory only mattes in that it becomes something you don't need to fiddle with while overclocking.
  • vol7ron - Friday, March 12, 2010 - link

    That's good to know.

    For some reason, I still have a hard time believing that modern systems are unaffected by such timings, not just CAS Latency, but the RAS-to-CAS delay, or even the tRAS. The memory bus may have increased, but the clock has basically stayed the same (266MHz and less) and the memory timings have gone up.

    If timings have gotten worst and the core clock has remained the same, then adjusting the clock/timings on those sticks today should have even a greater impact on today's sticks - this includes real life and benchmarks (the difference b/t the two is how likely your real life use will reach the benchmark scenario).

    ---

    What might even be nicer than this is to see if it's possible to fiddle with SSD RAM settings :) I'd like to see what kind of affect a 10% increase in clock speed, or 25% drop in timings would have! While I personally wouldn't mess with the stability of SSDs, this may be an option in the future, when I have spares laying around. As SSD controllers have become dependent on their RAM, I'd really like to see the impact this would have - perhaps the more common, cost-effective products on the market could surpass the 3Gbps threshold.

    I'd like to know what you think,
    vol7ron
  • drewintheav - Sunday, March 14, 2010 - link

    i heard that the i7 980x has dual qpi's and can run in a dual cpu configuration. like on an evga 270-gt-w555. can anybody confirm this?
  • gigi127 - Saturday, March 20, 2010 - link

    hello, I have also asked this question (I have a gulftown Q3QP 3.07 ghz) and I went to the intel site (http://ark.intel.com/Product.aspx?id=47932)">http://ark.intel.com/Product.aspx?id=47932) and the answer is that the x980 is not dual QPI,
    I contacted EVGA to ask if my processor is compatible with the new motherboard classified sr-2, and I was told that there is no problem in any case it is better EVGA contact for more details, I hope to be helpful.
  • bsoft16384 - Saturday, March 13, 2010 - link

    The reality is that in any modern CPU, with most programs, most memory requests are serviced by the cache. Once you get out of the cache you're totally screwed in any event.

    That's why Core 2 managed to be competitive with (and often faster than) Athlon 64 despite having a dramatically slower memory subsystem.

    Ironically, a design like Atom would benefit much more from faster memory than something like Nehalem. Atom has smaller caches and is in-order, so you get more capacity misses and the CPU can't order around them.

Log in

Don't have an account? Sign up now