Intel i5 750 Overclocking Advice/Help

[big4x4]

Hello all,

Been out of the overclocking game for a while and finally got a new setup. Have the following components:

i5 750
gigabyte p55a-ud3
4gb dd3
750 gb seagate
ocz 600w modxstream

So far, the highest stable OC i've gotten with stock volts is 3.3 + turbo. I just finished a 24hr session of prime95 @ 3.7 ghz + turbo (~3.9) @ 1.21 volts. Thinking that the best overall overclock for performance and efficiency is 3.6 ghz + turbo (~3.8) @ 1.68v. Have a couple questions though:
1. The computer seems to OC well but you really need to start pumping voltage in. I can do 4.0 without turbo but needs 1.31 volts. It seems like 3.6 ghz is the sweet spot since I only have to do a couple notches about stock volts and 3.7 plus turbo needs 1.225 in the bios for stability.
2. Is 24 hours of prime testing enough for general system stability? Haven't really used any other programs but OCCT. Any suggestions?
3. Do people normally keep turbo on? I'm a little confused about it and have heard that when only 2 cores are stressed, the multiplier jumps to 24x? Is that true? I know when I am stability testing, the multi is at 21x since all 4 cores are stressed.
4. I have all the energy saving features on but have yet to see the voltage decrease. The multiplier will jump down to 9 but isn't the voltage supposed to drop as well?

Thanks for any help and sorry about all the questions!

 

[lopri]

3.3 GHz + Turbo @stock seems alright to me. 3.7 GHz + Turbo with a minimal voltage increase is great. No need to pump voltages to achieve 4.0 GHz for 24/7 usage. 24 hours of Prime is way more than what I used to do to verify my overclocking. (Usually 4~8 hours) Though I've switched to LinX (Linpack) sometime last year. Some like Linpack over Prime and vice versa, but for me Linpack is more satisfying because I can easily set my goal (?) prior to test. (e.g. 50 loops using all available memory)

With regard to #4, make sure you have turned on SpeedStep and C1E (and C2E/C4E if present) in the BIOS. Check if Gigabyte has published newer BIOS.

 

[ScorcherDarkly]

OCing with turbo on is possible, you just need to be careful.

In order for Turbo mode to be enabled, the multiplier has to be set at 20. This is the multiplier the chip uses for 4 core ops (I think, it might use 21, but that doesn't make sense to me if it does). For 3 cores, the multiplier jumps to 21, and for 1 or 2 cores it jumps to 24. So at 160 BCLK, 4 cores is 3.2 ghz, 3 is 3.3, and 1 or 2 is 3.8.

So with that knowledge, setting your 4 core speed to 3.7 ghz would require BCLK of 185, which would make your 1-2 core speed 4.44 ghz. I hope you have water cooling. This is why you need to be careful.

If you're worried about efficiency, read this. The most efficient OC they could find was 160 BCLK with turbo enabled.

I think the only way you're going to see your voltage drop when the processor is idle is if you used the "auto" setting on vcore. Obviously you can't do this when OCing. I'm not certain about this, however, so someone correct me if I'm wrong.

 

[big4x4]

Thank you for all the replies! Very helpful information! Very interesting article. I have yet to see the 24x multiplier though. How can I get cpu-z to see it and/or test it for stability at 3.6ghz (which should be ~ 4.3 ghz....yikes!). I don't have water cooling but have not seen the temps go above 60 at full load for hours. Do you think I'll be ok at 3.6 + turbo enabled? Or should I drop it even farther with lower vcore?

 

[big4x4]

Update- Tried running prime on just one core and the multiplier still is only 21x

Also, it is interesting that my vcore does not change under load AT ALL. Setting the bios to 1.175 yields 1.152 in windows and will not change at all. Bumping it up one notch to 1.1825 yields 1.168 in windows and doesn't change. Funny how one little tick up make a bigger difference in windows!

Another thing I notice as I lower the voltage....why does it seem like core 3 always fails first? Is that normal?

 

[iluvdeal]

Try LinX instead of Prime95. My multiplier and vcore both clearly change when I run that.

 

[MrK6]

The voltage will only decrease if you keep the voltage on "Normal" and use the dynamic VID control (DVID) setting to adjust the voltage accordingly. For some reason, setting the voltage manually disables the voltage lowering. As far as why you can't get the 24x multiplier to work, IIRC, the multiplier adjust according to TDP, so if you're already high enough to surpass the 95W rating, it'll only give you 21x on load (this is just a guess though, someone please correct me if I'm wrong). FWIW, I doubt you'd have stability with two cores @ 4.3GHz using the same voltage that stabilizes four cores @ ~3.76GHz.

Core 3 is probably the weakest core in the CPU (weakest link). Either way, if one core isn't stable, the CPU isn't stable, so adjust your voltage accordingly.

 

[big4x4]

LinX really does a good job of finding errors faster than prime. And I do not think I am over 95w....if you look at that review posted of the i5 750, i'd guess I was in the area of 75-85. Mrk6, I am assuming you did away with Turbo correct? What volts you running at?

 

[MrK6]

I believe you must have turbo enabled on Gigabyte motherboards in order to use the 21x multiplier, so yes, I have turbo enabled. At 3.7GHz alone is probably enough to push you over TDP - rough estimate, I'd say you're at 105W. Even if you aren't, you're not overclocking to the 24x multiplier for dual core operations. It might be because you're not using the dynamic VID control, and so advanced power management isn't available (you have to downclock two cores to push the other two). That said, if you're going for the highest overclock with the least amount of voltage, the easiest way to go about it that I've found is to disable all power saving features (EIST, C1, C3/C6/C7), enable turbo, and crank it until your BCLK fails. The data in my sig is old, I've actually changed my overclock (more on that later), but for those settings I was running 190x21 @ 1.35V BIOS, ~1.34V real (Vdroop correction on). Because Intel specifies a maximum VTT of 1.21V, BCLK's are really limited (unless you don't care, in which case you can probably push these chips even further). I know my BCLK starts to get flaky at around 198ish, 200 won't stay stable for me. I tried testing higher to see if it was a fluke, but not too extensively. However, given that Bloomfields crap out at ~220 BCLK with ~1.35V on VTT, I'm not too optimistic (and given my smaller case and cooling solution, I'd be thermally limited anyway). The downside to this configuration is that it wastes a lot of power on idle (idle was ~123W from the wall).

All that said and done, I'm currently testing my rig at 3.6GHz, all power saving options on. Currently using ~1.2V load and lowering it slowly, with Vdroop correction off. The higher voltage at idle I believe keeps the processor stable while switching speeds, while Vdroop at load lowers the actual voltage going to the chip, bringing temps down (so Vdroop is actually beneficial here, who'dathunk). Basically, the problem isn't whether or not the processor is stable at load, it's whether or not it's stable which it's switching states. For example, I'd crash as soon as I stopped Prime95, even though it ran without error for 4 hours. The i5 750 is a different beast because it's power management is so aggressive.

 

[big4x4]

Awesome reply! Thanks for all the info. So you are pretty much running the same overclock I have meaning 180 x 20? 3.6 + turbo which means ~3.8 when it kicks to the 21x multi. For some reason, I am getting ZERO v-droop according to cpu-z. When I say zero, I mean IT STAYS CONSTANT! Don't know why, but I guess it kinda helps my oc a little lol. Just did 50 passes of LinX after bumping the vcore up one notch to finish that (it's a lot more stressful than prime....I was surprised!). My comp has yet to freeze yet which is surprising. None of the stress tests have locked up and have only come up with an error on one of the cores! That is with ALL of the power saving features on.

 

[MrK6]

Awesome reply! Thanks for all the info. So you are pretty much running the same overclock I have meaning 180 x 20? 3.6 + turbo which means ~3.8 when it kicks to the 21x multi. For some reason, I am getting ZERO v-droop according to cpu-z. When I say zero, I mean IT STAYS CONSTANT! Don't know why, but I guess it kinda helps my oc a little lol. Just did 50 passes of LinX after bumping the vcore up one notch to finish that (it's a lot more stressful than prime....I was surprised!). My comp has yet to freeze yet which is surprising. None of the stress tests have locked up and have only come up with an error on one of the cores! That is with ALL of the power saving features on.

Well no, now that I'm at 180x20 I have turbo off, simply for the reason that if I wanted 3.8, I would run 190x20. It's just another (very significant) confounding factor that I enable only when I need to - i.e. I need the 21x multiplier. As I said, I believe the problem isn't when the CPU is idle or at load, but when it switches between the two. I know 1.225V is relatively stable for 3.8GHz if I lock the voltage and disable power saving, but I'm currently testing 1.25V (and hopefully lowering it) for 3.6GHz just because there's that much of an offset. I'll keep tinkering around with it when I have the time. As long as you're stable, keep pushing it, I look forward to your results. Also, sorry if you already knew this, but just to clarify voltage should stay constant even with Vdroop correction (LLC) disabled, all it is is that there's a discrepancy between what you set in the BIOS and what it actually applies at load.

 

[big4x4]

Thanks for the response. I look forward to hearing about what your OC ends up being as well! Just got finished test 100 loops of LinX @ 3.7 (175 x 21) with only 1.152v on CPU-z. Pretty sure I'll leave it at that for today but tomorrow I am going to try to tweak to get the voltage down a little if I can. This is with all the power saving features on! I may try for 3.8 tomorrow again (180 x 21) but didn't have luck with stability in LinX under 1.2v. I'll try some different settings and get back. But I am happy with ~3.7 at 1.152 volts for now!

Regarding vdroop....I knew that cpu-z always reads lower in the bios. On ever system i've built besides this one, the voltage ALWAYS dropped under load according to cpu-z so that's why I asked.

 

[bob5568]

@ MRK6> Not sure how you reach the conclusion that vdroop = 0. Nothing I've read supports that assertion. My computer shows a clear vdroop as well as vdrop, which you do recognize. Have you read this article: http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=3184&p=5 ? My vdroop (difference in Vcore at load vs idle) is non-linear, but hovers around .02 to .03 v with Vcore (bios) around 1.28v. Vdroop gets bigger as the bios vcore setting gets bigger.

 

[MrK6]

@ MRK6> Not sure how you reach the conclusion that vdroop = 0. Nothing I've read supports that assertion. My computer shows a clear vdroop as well as vdrop, which you do recognize. Have you read this article: http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=3184&p=5 ? My vdroop (difference in Vcore at load vs idle) is non-linear, but hovers around .02 to .03 v with Vcore (bios) around 1.28v. Vdroop gets bigger as the bios vcore setting gets bigger.

I have no idea what you're trying to get across here. Could you quote which part wasn't clear to you or you are not agreeing with?

 

[bob5568]

Also, sorry if you already knew this, but just to clarify voltage should stay constant even with Vdroop correction (LLC) disabled, all it is is that there's a discrepancy between what you set in the BIOS and what it actually applies at load.

Sorry for confusion...I took this statement to mean you think vdroop = 0 (because you say that voltage should remain constant).

 

[Kenmitch]

I'm kinda following your thread as Im building similar setup.

Just got my CPU last night and finished build around 12:15 this am. I didn't get much time to play around with it yet...but it looks like I gotta 4ghz + chip

I gotta sit back and do some reading tho as I'm new to the platform and bios looks kinda greekish compared to s775.

I was more concerned with heat as it seems like these chips run hot at high clock speeds. But seems like the water cooling handled that for me. I did notice when double checking for leaks befoe I shut her down that I forgot to plug the fan in on my second rad.

Good luck on your overclocking!

I'll spin off my own thread once I get to the fun stuff

 

[big4x4]

Kemmitch,

Let me know what your OC ends up being and at what volts! Still at 3.7 ghz @ 1.152v. Going any lower would error out LinX within an hour or so!

 

[Munky]

There's a difference between turbo and "turbo." For example on my Asus board, "turbo" simply refers to the old turbo method, where all cores can work at 1 multi higher than default. The new method of dynamically raising the multi depending on how many cores are loaded is called "c-state tech." You can enable one method without the other.

Also, nothing wrong with keeping the voltage setting at "auto" for a medium 24/7 OC. Keeping the multi on Auto and the voltage on Auto will drop your cpu to lower voltage and multi when it's idling. This is a great option for typical 24/7 use.

From my experience, an OC using the dynamic turbo mode can perform worse than a comparable static turbo OC. I ran some game tests (Farcry 2), where the static method clocks all cores to 3.54ghz under load, whereas the dynamic method clocks from 3.3 - 3.9ghz depending on how many cores are loaded, and the static method returned faster results. If you're only doing single-threaded work, like mp3 encoding, then the dynamic mode can potentially be faster though.

 

[Kenmitch]

Kemmitch,

Let me know what your OC ends up being and at what volts! Still at 3.7 ghz @ 1.152v. Going any lower would error out LinX within an hour or so!

I'll do that for ya!

My plan is to try and leave all chip features enabled. Just gotta do some homework on it.

I'm thinking at 4ghz vcore is gonna be around 1.30 or so but on water it should be fine.

 

[big4x4]

So 4.0 as in 200 x 20? Or are you talking about 190 x 21 w/ turbo? But good luck on getting it stable with all the features on! Definitely the most finicky oc i've done!

 

[MrK6]

Sorry for confusion...I took this statement to mean you think vdroop = 0 (because you say that voltage should remain constant).

Ah, no worries. I was pointing out that while in either case your voltage remains at a constant value (it'd have to, otherwise you'd have major issues with stability), that constant value is lower and further away from the BIOS set Vcore than if one enables LLC. For instance, you set 1.35V Vcore in the BIOS, with LLC off you might load at a consistent 1.31V (within normal variation), but with it on you might load at 1.34V, etc.

 

[Kenmitch]

So 4.0 as in 200 x 20? Or are you talking about 190 x 21 w/ turbo? But good luck on getting it stable with all the features on! Definitely the most finicky oc i've done!

Last night is kinda a blur as I got less than 4yrs sleep and am currently still at work.

I kinda booted up at 4ghz and did some quick test's to check temps.

I then tried for more 220x19 posted got into windows. Did some intel burn test to check temps. Was not stable but didn't have time to try to stabalize it as it was about 1:15 am by then. All energy features are enabled as far as I can tell. Both multiplier and voltage are fluctuating.

My goal is to get 4ghz stable. I'm not sure if it'll be whooped tonight or not. But I'm gonna give it a shot. I'm kinda hoping it takes some tinkering around as I upgraded for only the challenge

 

[big4x4]

Word of advice.....just disable the turbo feature! It will save you a lot of headache when overclocking lol. Good luck and make sure to post back!

 

[bob5568]

Ah, no worries. I was pointing out that while in either case your voltage remains at a constant value (it'd have to, otherwise you'd have major issues with stability), that constant value is lower and further away from the BIOS set Vcore than if one enables LLC. For instance, you set 1.35V Vcore in the BIOS, with LLC off you might load at a consistent 1.31V (within normal variation), but with it on you might load at 1.34V, etc.

Ahh, you were talking about 100% loaded voltage being constant. Thats what I missed. Thanks.

 

[VirtualLarry]

Last night is kinda a blur as I got less than 4yrs sleep and am currently still at work.

Regular rip van winkle, I see.