The Problem with Overclocking
Posted on 21st Dec 2009 at 10:50 by Clive Webster with 20 comments
For some completely unknown reason (alright, just one I can’t tell you yet), I’ve spent most of last week testing and overclocking CPUs. Because of the way the magazine deadlines work in the run-up to Christmas, deadlines are tight, but the urge to get the very last MHz from a CPU still bit me pretty hard.
One of the CPUs I was testing was a Core 2 Duo E7400 – we’d previously seen that the E7300 could overclock from 2.66GHz to 4.2GHz, and I was sure I could 4.3GHz out of it. In fact, I initially tried for 4.4GHz but just couldn’t get the second core stable under Prime. Nothing I tried could get that second core to play ball – the worker thread on Prime95 crashed out the moment I kicked in the smallfft test.
Normally we head off home at 6pm (we start at 10am, so it’s still an eight hour day) but here I was at ten to eight in the evening tapping numbers into my calculator, determined to get 4.3GHz out of the chip. I was played with the GTL for the second core, trying various combinations of PLL and vcore and generally trying everything I could to get that troublesome second core to work. Finally, after many restarts and failed attempts, I conceded that 4.3GHz wasn’t possible.
I quickly verified that 4.2GHz was fine, and, glancing at my watch to see it was now past 8pm, I tried to get the bloody CPU as close to 4.3GHz as possible. Why? Why did I do that? What difference was it going to make to any benchmark – let alone the review itself – if the CPU was clocked at 4.25GHz or 4.28GHz? What was I going to prove? I took me five or six further failed attempts to realise that actually the CPU wasn’t going to go any higher and I should settle for 4.25GHz.
The weird thing is, a few of years ago I’d have been thoroughly chuffed with a 52 per cent overclock for two hours of BIOS fiddling. The other night it just felt like a failure… This is just one of the many problems with overclocking, but my sleep-deprived, blood-shot left eye makes this particular issue the most annoying. Where do you draw the line between a worthwhile amount of performance gains for the time you've invested?
One of the CPUs I was testing was a Core 2 Duo E7400 – we’d previously seen that the E7300 could overclock from 2.66GHz to 4.2GHz, and I was sure I could 4.3GHz out of it. In fact, I initially tried for 4.4GHz but just couldn’t get the second core stable under Prime. Nothing I tried could get that second core to play ball – the worker thread on Prime95 crashed out the moment I kicked in the smallfft test.
Normally we head off home at 6pm (we start at 10am, so it’s still an eight hour day) but here I was at ten to eight in the evening tapping numbers into my calculator, determined to get 4.3GHz out of the chip. I was played with the GTL for the second core, trying various combinations of PLL and vcore and generally trying everything I could to get that troublesome second core to work. Finally, after many restarts and failed attempts, I conceded that 4.3GHz wasn’t possible.
I quickly verified that 4.2GHz was fine, and, glancing at my watch to see it was now past 8pm, I tried to get the bloody CPU as close to 4.3GHz as possible. Why? Why did I do that? What difference was it going to make to any benchmark – let alone the review itself – if the CPU was clocked at 4.25GHz or 4.28GHz? What was I going to prove? I took me five or six further failed attempts to realise that actually the CPU wasn’t going to go any higher and I should settle for 4.25GHz.
The weird thing is, a few of years ago I’d have been thoroughly chuffed with a 52 per cent overclock for two hours of BIOS fiddling. The other night it just felt like a failure… This is just one of the many problems with overclocking, but my sleep-deprived, blood-shot left eye makes this particular issue the most annoying. Where do you draw the line between a worthwhile amount of performance gains for the time you've invested?
20 Comments
Discuss in the forums ReplyThat's the reason I don't do case mods. There are some really talented case modders out there but I personally feel that life is too short to spend modding a case to perfection (unless of course the end result wins me fame and fortune).
And, going back to overclocking... I have had an awful few weeks since I ditched my 3.84ghz overclocked Q6600 system in favour of i7 860 on EVGA FTW P55 and GSkill 1600mhz ram. Only after I made the change did I read forum comments across the web that the 860 isn't really that great at overclocking past 4ghz. Most people don't even get to 4.2ghz. Once you get past 4ghz you need massive voltage increase. I am not sure at this point that I made the right choice going to P55 platform.
My original plan was X58/920 but I was lured by the lower power consumption of the 860 and also the fact that the EVGA FTW has mounting holes for my skt 775 waterblock as well.
My main concern is that 1156 will be neglected and all the best cpu releases will be for skt 1366.
Just the envy speaking here ;)
Totally agree about the MB. It's vital to have a really well built board to be able to cope with the extra signalling speed and volts demanded by an OC.
what defines an overclocker? the will to squeeze every last drop of performance by tweaking BIOS settings (or jumper pins in the old days)
sure the time spent might not be worth it in terms of performance, but think of the achievement you feel once you have found the absolute maximum of your chip?
i have yet to find my i7 860's max clock speed, and i am really looking forward to do so. it would be a joy to tweak the new architecture and experiment with the settings.
on the other hand, 24/7 overclock is more important, and with the wear level in mind, a slightly lower overclock should be used. for the e7400 i'd say 4Ghz.
@feathers. the LGA1156 platform won't be left behind. it will cover the high-end down to the bottom end, no reason to get scared :)
also, 4Ghz on 860 is very easy as long as you have a lucky chip. low VID of 1.12v helped to get my i7 860 to 4Ghz with only 1.3v, same as other i7 920 chips.
4ghz has indeed been easy on 860. It was only let down by unstable BIOS on EVGA. That has now been rectified with A39 bios update. I don't think I have a lucky chip but I should spend some time to see if I can reduce the volts a little now. VTT = 1.36. Vcore 1.35 in windows. From what I've read, my volts are quite common for 860. Fudzilla's review of the EVGA FTW used similar volts to get to 4ghz and they also had problems getting 4.2ghz without massive voltage raise.
:)
i wouldn't say that's a large voltage increase. however from 3.8Ghz to 4Ghz does need a large-ish voltage increase, from 1.23v jump to 1.3v to pass my extensive stress tests.
and that also comes back to problems with overclocking. the 1.23v wasn't tested extensively as i just tested 1.2v found to be stable enough, and jacked up a bit on voltages just to be safe. very quick 30min of work. but to get 4Ghz stable without blowing up the CPU when running LinX (Lin pack GUI) i had to spend more than 2 hours just tweaking it and then there's the electricity bill for stress testing it over 4 nights. does the 200Mhz increase matter? is it worth the effort? YES for both question as i was able to achieve my goal of 4Ghz. on 4th morning, when i saw the stress test was still going without errors, i was so pleased.
i wouldn't say that's a large voltage increase. however from 3.8Ghz to 4Ghz does need a large-ish voltage increase, from 1.23v jump to 1.3v to pass my extensive stress tests.
and that also comes back to problems with overclocking. the 1.23v wasn't tested extensively as i just tested 1.2v found to be stable enough, and jacked up a bit on voltages just to be safe. very quick 30min of work. but to get 4Ghz stable without blowing up the CPU when running LinX (Lin pack GUI) i had to spend more than 2 hours just tweaking it and then there's the electricity bill for stress testing it over 4 nights. does the 200Mhz increase matter? is it worth the effort? YES for both question as i was able to achieve my goal of 4Ghz. on 4th morning, when i saw the stress test was still going without errors, i was so pleased.
It was prime stable at 3.69ghz....
thats an annoying number...
but come rain or shine it would not be prime happy past that.
Which is why I sold it to my brother and bought a 965 C3 ^__^
I think nowadays the spot that everyone wants if 4.0Ghz, because its well beyond the range of anything stock speed and when you say 4.xxGhz people know its something special. anything 3.xx its just not quite as good. Which is what makes me ecstatic about my 4.02GHz Prime stable OC. (oh and it'll post up to 4.64GHz too ^__^ )
wuyanxu is further proof of this theory ^_^ and same on the cpu cooler, it really is a beast. but with the fan set to voltage not pwm its silent too
very nice OC at 4.6Ghz.
knowing i can get rock-solid stable 3.5 out of a 2.8ghz chip, under a $20 air cooler, on a budget p43 mobo with OEM ddr2 800 ... for as little as i spent to put this rig together, is amazing to me.
i like it so much, i dug up another e7400, and plan on building a htpc around it and some found / donated parts... i'm curious to see how it handles an old 650i board. probably won't be overclocking the movie box though.
I think you ought to look at the number you started out with and the number you finished with and ask yourself, "What did I set out to do in the beginning?" and I think you may find the answer to be "To push this processor to it's limit" and you have found it's limit (for the time being).
leading to my 600mhz increase in my E8400 when I full well know that I can hit 4ghz.
the e7400 has a smaller cache.. it's a oc beast on a budget =] perfect chip really- and from a gaming point of view it's not about the cache as it is the speed
thing is if you plan on any kind of longevity (2yrs+).. you need to keep it under thermal specs 74.1C - and I go a step further and keep it at or under the max voltage spec (according to intel 1.3625v).. I always looked at it like who would know better about the chips limits and electromigration than intel's own engineers
http://processorfinder.intel.com/details.aspx?sSpec=SLGQ8
your pretty much guaranteed a good life on the chip