i5 750 Overclock causing performance reduction...

[beastly_divizion]

Ok, So I'm FAIRLY knowledgable regarding overclocking - Been doing it for a long time now - Started way back, When the only way to overclock required a skilled solderer.

Anyway, Here's the specs for the problem machine:

Aerocool BX-500 Ultimate Case with 10 fans, All stock fans replaced with Noctua
Gigabyte GA-P55A-UD3R Motherboard - f8 Firmware version
Intel i5 750 CPU with Noctua NH-U12P Cooler using both fans in Push-Pull
4GB Corsair Platinum DDR3 RAM 9-9-9-24 @ 1333MHz
WD Caviar Black SATA3 HD 1TB
Sapphire Radeon 5870 (NOT VAPOR-X) Reference Board
Antec Quattro 1200w PSU


Here's the issue:
I can reach a stable overclock of 4.25GHz without exceeding 65 degrees celsius under Prime95 and IntelBurnTest load. Ran for 3hrs straight and still stable at the end.

However, I noticed that the realworld performance of the overclock was shoddy, Especially when gaming. Even when just running Windows, Opening programs took longer. Gaming sees jerky frame rates, Even though, According to MSI Afterburner, The GPU Usage never peaks above 70% and CPU usage never reaches above 65% on all cores, And the longer load times, So I dropped the O.C. Back to 3.25GHz and the load times, Dropped drastically and frame rate remains stable.

My question is this: Why and under what circumstances, Would overclocking by more than 1GHz, Result in a reduction in performance and overclocking by 600MHz result in an increase in performance?

Is there some kind of overclock ceiling, Where passing that starts to degrade performance?

BTW to achieve the 4.25GHz OC, I bumped my CPU voltage up to 1.38125, Multiplier to 20x, Baseclock to 212.

[burebista]

Originally Posted by beastly_divizion

I can reach a stable overclock of 4.25GHz without exceeding 65 degrees celsius under Prime95 and IntelBurnTest load. Ran for 3hrs straight and still stable at the end.

Those are core temperatures measured by RealTemp/CoreTemp?

[Rison]

The only thing I can think of, is if your temp monitoring is off. Some programs can be off as much as 10c.. and your CPU could throttle, which throws any performance gain right out the window.
It would still "pass" prime95 and intelburntest when it throttles.

I'd drop the overclock back to 4.1Ghz, then drop the vcore as low as it would go stable.. and see if that makes a difference.

[beastly_divizion]

Originally Posted by Rison

The only thing I can think of, is if your temp monitoring is off. Some programs can be off as much as 10c.. and your CPU could throttle, which throws any performance gain right out the window.
It would still "pass" prime95 and intelburntest when it throttles.

I'd drop the overclock back to 4.1Ghz, then drop the vcore as low as it would go stable.. and see if that makes a difference.

Temps are measured by RealTemp, Which was recommended to me, During the Prime95 load...Could not measure during IntelBurnTest as it pretty much used all of the available resources.

Is it a reliable temp monitor or should I try another temp monitor?

[sswilson]

Is it possible that your bus is overtaxed at that speed when running real world apps but not during stress testing?

I haven't OC'd on the i7/15 platform so I can only comment on observations on other platforms, but I've found that a fully stable (12+ hrs Prime) OC would behave like you've described if the NB voltage was set too low....

Not something that can be measured or quantified, but I often find that a slight bump to NB V (after it's been stress tested) provides me with a much snappier and smoother OS. In XP, the visual cue was watching the scrolling bars during load.... if they seemed to hang up (and recover) mid-load, a bump to NB V would more often than not make that "bump" go away and resulted in a better OS experience.

YRMV.

[beastly_divizion]

How slight are we talking here?

The motherboard in this setup only allows for somewhere in the vicinity of 0.00125 volt increments. If we're talking less than that, I guess I'll have to reduce the overclock somewhat.

Unfortunately, I don't run XP anymore, So I can't really use that test. I know that the OC is sufficiently stable at 4.25GHz, It's just the performance issue.

I'll test your NB Voltage suggestion now and report back.

Out of curiosity, Is it even remotely possible that there's a max overclock ceiling, Where, No matter what changes you make to any voltages or settings, You don't see any measurable performance increase?

[sswilson]

Originally Posted by beastly_divizion

How slight are we talking here?

The motherboard in this setup only allows for somewhere in the vicinity of 0.00125 volt increments. If we're talking less than that, I guess I'll have to reduce the overclock somewhat.

Unfortunately, I don't run XP anymore, So I can't really use that test. I know that the OC is sufficiently stable at 4.25GHz, It's just the performance issue.

I'll test your NB Voltage suggestion now and report back.

Out of curiosity, Is it even remotely possible that there's a max overclock ceiling, Where, No matter what changes you make to any voltages or settings, You don't see any measurable performance increase?

Not that I can think of when it comes to CPU overclocking (other than the NB issue I've already mentioned), but memory because of the secondary timings all being dependant on/triggered by each other would have a "sweet spot".

[ForTheGG]

Just a thought, when overclocking with BLCK you're memory also gets overclocked. If your timings aren't set in the BIOS then you may be using something crazy like DDR3-1700 @ 10-10-10-30, which may have a negative impact on your overall system performance.