My i5 760 Overclocking adventure!?

[Dward]

Hi guys, I am a long time reader of AnandTech. Great website, so I thought it was about time to sign up here and share my overclocking for views and help.

My previous system was a E6550 which I had overclocked to 3Ghz. So I'm not new to overclocking but I'm certainly no expert!

I am stuck in two minds with what to do next with my system but before I get in to that, here are the details...

My current system which I bought in August 2010 is:
i5 760
Gigabyte GA-P55A-UD3
G.Skill 4GB (2x2GB) kit, PC-12800 (1600Mhz), 6-8-6-24-2N
Corsair CMPSU-650HX 650W ATX Modular PSU
Gainward GTX 460 GS-GLH
1TB Western Digital Black Caviar
Antec 300 Case
2xScythe S-FLEX 120mm Fan 1200RPM (Front of case)
Xigmatek Dark Knight Revision 3 CPU Heatsink & Cooler

I bought this to be able to run games at 1680x1050 on high detail. I currently only play BC2 but with games like Crysis 2, Dues Ex & BF3 on the horizon I needed a good quality and reasonably affordable PC to do this.

It did not take me long for the overclocking bug to bite, so off I went in search of higher clock speeds and fast memory timings. (As at stock the memory only runs at 1333Mhz and CL9)

I wanted it to still run at low clock speeds during windows but wind up to super speeds when all 4 cores are loaded. So I left the power saving options on and turned OFF Turbo.

My current stable overclock (after quite a few lock ups and BSOD's) is:
160x9 = 1440Mhz @ 0.864V at idle &
160x21 = 3360Mhz @ 1.088V at load.
Memory @ 1600Mhz (160x10 SDP) 6-8-6-24-2N

I am fairly happy with this, my temps don't go any higher than ~50C. Even when I run Intel Burn Test for hours (OCCT Version).

Here are the BIOS settings with this current overclock: (Does everything look ok??)

http://s140.photobucket.com/albums/r6/dward_87/New Computer/Overclocking/

Here is a picture of the voltages and temps under load and idle:

This setup runs stable in windows, BC2 & OCCT (both OCCT test and Intel Burn Test)

Is there any other programs I should run to test the stability? Prime95? LinX?

I do get one slight occurring thing happen with my CPU Fan that is a little worrying, every now and then during Windows use (no load), the CPU fan spins up to ~2000rpm for about 3 seconds then goes back to around 1500rpm. My temps don't go up nor does my voltages. Should this be a concern?

My other question and I guess the BIG ONE, is what to do next? Should I aim to get it stable with TURBO on? (I have had a couple BSOD's when I have tried to this in the past). Or should I aim to increase the BCLK and leave turbo off?

Any other thoughts would be greatly appreciated.

Thanks

 

[brencat]

Hi Dward, welcome to AT forums. You have a nice system there and that chip is a great overclocker.

First off, have a look at the below *short* OCing guide which is the one I used successfully and is great at explaining all the moving parts involved:

http://www.techreaction.net/2010/09/07/3-step-overclocking-guide-lynnfield/2/

Next, take a look at post #7 of this thread, in which I discussed the OCing steps incorporated in the article above:

http://forums.anandtech.com/showthread.php?t=2123481&highlight=

In terms of final stability testing once you think your OC is set where you want it, run 100 passes LinX (Linpack) set to use "All" memory, followed by 3 hrs OCCT on "High Data Set", followed by at least 12 hrs Prime 95 (sometimes one of your cores will fail ~ the 10 hr mark if vcore is a notch too low). If you pass that, you're good to go, IMO.

Mine: http://img835.imageshack.us/img835/7519/i576038ghzloaded.png

 

[Dward]

Wow great read! It seems like I have only scratched the surface!

Ok so I think the plan for me now is to stress test this current system with LinX and Prime95 to make sure this is stable. Then I'll come back and reach higher speeds!!

Cheers

 

[Dward]

Ok I have not had time to do so many Linpack passes, only 20, which returned no errors.

While I was at work today I left Prime95 running, here are the results

I have passed OCCT on High Data Set previously for over 8 hours too with out any errors.

So I would presume my system is all good. (Will let Prime95 run for over 12 hours in the next few days).

Should I be concerned about the random spinning up off the CPU fan every now and then in windows? I found a BIOS setting called CPU Smart Fan Mode which had either "AUTO", "PWM" and "Voltage"

I tried PWM and it still spins up to 2000rpm in windows, I then tried Voltage which makes the CPU fan only spin at ~600rpm on idle and ~1400rpm on load, which I thought was way too low (too risky).

Thoughts on this please?

 

[faxon]

i personally am a silence nut. what i do is set the fan to 100% all the time, and if it's to loud then i use something quieter. right now im using low speed yate loons on a triple 120 radiator watercooling the system i use as my primary, but before that i used a prolimatech megahalems with push/pull low speed yates. system is in sig. if i were you, i would just use OCCT (and occt linpack) to test your system vs P95. P95 uses a very similar algorithm to OCCT, but OCCT seems to find instabilities about 10 times faster for some reason, while generating similar temperatures. linpack is the ultimate end all to testing, and if your system is able to pass it i'd say its gonna be fine under most any conditions (though with sandy there seems to be some exceptions). also with that chip @ that voltage i'd push for at least 4GHz if i were you

 

[Dward]

Thanks Faxon. I'm definattly going to leave my current CPU Heatsink/Fan. It does a great job in keeping the temps low. Plus I don't want to go through the process of installing another once, since I got this one installed by a pro. (as proven by the great install, see pics)

As for pushing it more, I'm in two minds!!! Keep it as is as I don't NEED more Ghz, or see what this baby can do?!!

 

[perdomot]

I upgraded to a i5-760 at the end of Nov due to a Black Friday sale at Newegg and chose this cpu because according to most reviews, it OCs to 4Ghz easily and with little voltage. Unfortunately, the mobo I got died and I've been using an H55 mobo since so my OC is kinda limited but I can't complain about the 3.6Ghz I've achieved. All I had to do was change my multi to 20, Bclk to 180, disable turbo-c1e-eist but left c3/c6on and set vcore to 1.20. Also adjusted ram multiplier to keep it in spec. This review video gives you great info so you might want to see how far you can take your rig:
http://www.youtube.com/watch?v=AeF4NsPpEq0

 

[brencat]

As for pushing it more, I'm in two minds!!! Keep it as is as I don't NEED more Ghz, or see what this baby can do?!!

Well, how about somewhere in the middle? Assuming you've found the max Bclk (or hit 200) with no more than 1.30 - 1.35v QPI/VTT, you should be able to get 200 x 18 (3.6ghz) with stock vcore. From there, your chip will likely start needing more.

For me, I'm using a 92mm cooler so I never intended to get to 4.0ghz. I was prepared to accept 3.5 - 3.6ghz when I bought the chip but I got 200 Bclk pretty darn easily with only 1.25v QPI/VTT, which allowed me to use 200 x 19 for 3.8ghz so I was thrilled!

But just to go from 3.6ghz on stock vcore to 3.8ghz required an additional +0.08125v. I decided the voltage required to get to 4.0ghz was not worth it for me, and also given the temps -- which are well within acceptable now, in winter, but allowing for ~ 7c difference in my room temp come summer factored into my decision to stop there.

 

[Dward]

Ok yeah I like the sound of 3.6Ghz running 200Bclk and the multiplier at x18. This way I can still have the memory run at 1600Mhz and CL6 timings (8xSPD).

So! Can you please list the steps I should take to get there? Should I leave my Vcore set to "Normal" and my Dynamic Vcore to "+0.0125" like it is at the moment? And just fiddle with the QPI/VTT Voltage...

Thanks again

EDIT: Just read your post (#7 http://forums.anandtech.com/showthread.php?t=2123481&highlight=) again. It makes more sense now! Will give that a go and see how I go. So my plan is:

-Leave CPU Vcore on normal
-Put Dvid back to normal (or +0.0, what ever the setting is called)
-Load-Line calibration leave on Enabled
-Leave memory voltage at 1.6
-Change QPI/VTT to 1.2V
-QPI Clock Ratio to 32x ???
-Reduce CPU multiplier to 9x
-Leave BCLK at 160Mhz ???

Then stress test using 20 passes of LinPack.

-Increase BCLK by 10Mhz increments, if LinPack fails, increase QPI/VTT by 0.02V until I pass LinPack & OCCT with a QPI/VTT voltage no greater than 1.3V?

Then increase the CPU multiplier by 1 and stress test each time with LinPack & OCCT. If these fail, increase Dynamic Vcore.

Hopefully reach 200x18 @ 3.6Ghz!

How does that sound?

 

[perdomot]

I raised my vcore to 1.25 because when I watched Flash video, anything less would freeze the PC. Not always but any lower voltage would inevitably cause the freeze.

 

[RussianSensation]

So! Can you please list the steps I should take to get there? Should I leave my Vcore set to "Normal" and my Dynamic Vcore to "+0.0125" like it is at the moment? And just fiddle with the QPI/VTT Voltage...

Perfect. Leave the DVID setting and EIST/C3-C6 enabled. Thing should still work. Increase QPI/Vtt to 1.25V. Increase DVID (dynamic Vcore) to result in load CPU voltage of 1.312V or so. Don't worry about syncing memory to run at DDR3-1600. Core i series architecture scales with clock speeds and doesn't really care for memory bandwidth.

Try 200x19 OR 190x20. If the system is not stable, try raising the Vcore a little bit and/or QPI/Vtt to say 1.28V. If things are smooth, reenable C1E and see if everything is smooth.

 

[Dward]

Hmmm just BSOD'd playing Mass Effect 1 :/

Let's get in to this overclock more to hopefully make it more stable. If that is possible lol

 

[brencat]

Ok yeah I like the sound of 3.6Ghz running 200Bclk and the multiplier at x18. This way I can still have the memory run at 1600Mhz and CL6 timings (8xSPD).

EDIT: Just read your post (#7 http://forums.anandtech.com/showthread.php?t=2123481&highlight=) again. It makes more sense now! Will give that a go and see how I go. So my plan is:

-Leave CPU Vcore on normal
-Put Dvid back to normal (or +0.0, what ever the setting is called)
-Load-Line calibration leave on Enabled
-Leave memory voltage at 1.6
-Change QPI/VTT to 1.2V
-QPI Clock Ratio to 32x ???
-Reduce CPU multiplier to 9x
-Leave BCLK at 160Mhz ???

Then stress test using 20 passes of LinPack.

-Increase BCLK by 10Mhz increments, if LinPack fails, increase QPI/VTT by 0.02V until I pass LinPack & OCCT with a QPI/VTT voltage no greater than 1.3V?

Then increase the CPU multiplier by 1 and stress test each time with LinPack & OCCT. If these fail, increase Dynamic Vcore.

Hopefully reach 200x18 @ 3.6Ghz!

How does that sound?

Yep, you appear to understand perfectly. Now get it done soldier... :thumbsup:

 

[brencat]

Perfect. Leave the DVID setting and EIST/C3-C6 enabled. Thing should still work. Increase QPI/Vtt to 1.25V. Increase DVID (dynamic Vcore) to result in load CPU voltage of 1.312V or so. Don't worry about syncing memory to run at DDR3-1600. Core i series architecture scales with clock speeds and doesn't really care for memory bandwidth.

Try 200x19 OR 190x20. If the system is not stable, try raising the Vcore a little bit and/or QPI/Vtt to say 1.28V. If things are smooth, reenable C1E and see if everything is smooth.

RS's advice is always among the best on AT OC forums, but in this case I will only suggest to you to use the absolute minimum additional dynamic vcore increase that YOU need for YOUR overclock target.

That kind of voltage would be appropriate if you were going for 4.0ghz or greater or if you had the older i5-750 which had higher starting VID. But to do 3.6 - 3.8ghz on *most* i5-760s don't need anywhere near 1.20v vcore let alone 1.30v.

The mistake many people make/have made with these i5 chips is that they don't use enough QPI/VTT voltage, so they have to make up for it with way more vcore than is necessary, which raises temps and stress. Assuming you follow our suggestions from earlier, you will likely need a lot less vcore to accomplish your goal.

 

[perdomot]

RS's advice is always among the best on AT OC forums, but in this case I will only suggest to you to use the absolute minimum additional dynamic vcore increase that YOU need for YOUR overclock target.

That kind of voltage would be appropriate if you were going for 4.0ghz or greater or if you had the older i5-750 which had higher starting VID. But to do 3.6 - 3.8ghz on *most* i5-760s don't need anywhere near 1.20v vcore let alone 1.30v.

The mistake many people make/have made with these i5 chips is that they don't use enough QPI/VTT voltage, so they have to make up for it with way more vcore than is necessary, which raises temps and stress. Assuming you follow our suggestions from earlier, you will likely need a lot less vcore to accomplish your goal.

My 3.6Ghz OC needed more than the 1.20vcore I was giving it to be totally stable. I followed 3D Overclock's recommended 1.25 setting and it works fine.

 

[perdomot]

Hmmm just BSOD'd playing Mass Effect 1 :/

Let's get in to this overclock more to hopefully make it more stable. If that is possible lol

What settings were you using? Be sure to set your ram properly and manually set the timings.

 

[yepp]

I'm just using the standard CPU Vcore settings, Load-Line Calibration(LLC) disabled(on the fence with enabling this). Did a quick run at 4Ghz, my i5-760 needed around 1.26v during vdroop when loaded to past a quick run of LinX. That was around 1.33v idle and 1.375vid selected in BIOS, had the VTT at 1.31v IIRC.

Have not messed around with the Dynamic Vcore(DVID), does it work well with and without LLC?

So with DVID the CPU will be supplied stock volts then increase volts when loaded. Or would it be supplied the increase volts even when idle? Is it wise to use this at high OC or best to use it during mild OC?

 

[brencat]

So with DVID the CPU will be supplied stock volts then increase volts when loaded. Or would it be supplied the increase volts even when idle? Is it wise to use this at high OC or best to use it during mild OC?

Yes, dynamic vcore is a permanent offset to the "Normal" vcore so yes, your chip will be supplied base vid + whatever you choose for dynamic offset when at idle with EIST/C1E on.

LLC also adds ~ 0.04-0.05v to vid but only when at load to prevent vdroop.

Example: I have LLC on and EIST on. My CPU loads 3.8ghz @ 1.12v. When at idle (9 x 200 = 1800mhz), CPUZ shows vcore as 0.98v. The differential is +0.08125v dynamic vcore and 0.05v from LLC.

 

[Dward]

Wow thanks guys! Ok I've started!

Managed to get to 200x9 @ ~0.848 "Normal Vcore" (have not touched that yet) & 1.25 QPI/VTT. Memory @ 1600Mhz 6-8-6-24 timings also

It would not boot with 1.23, so I upped it to 1.25 and she booted!! Now on to LinPack with "ALL" memory.

 

[Dward]

Ok LinPack Passed! What multiplier should I start at? 15?

 

[brencat]

Ok LinPack Passed! What multiplier should I start at? 15?

Awesome job so far. Let's make sure though ...did you do 50 passes Linpack at this juncture or just 20? And did you also run OCCT for 1 hour? You want to be sure 200 Bclk is decently stable before moving onto the final step. Otherwise if something crashes during the last phase of multiplier raising/vcore raising, you won't be certain if it's insufficient QPI/VTT related to Bclk instability, or insufficient vcore related to core speed.

 

[Dward]

Awesome job so far. Let's make sure though ...did you do 50 passes Linpack at this juncture or just 20? And did you also run OCCT for 1 hour? You want to be sure 200 Bclk is decently stable before moving onto the final step. Otherwise if something crashes during the last phase of multiplier raising/vcore raising, you won't be certain if it's insufficient QPI/VTT related to Bclk instability, or insufficient vcore related to core speed.

*sigh*, only 20 passes at this stage!

Ok will do some OCCT High Data set for an hour and some more "Intel Burn Test", hell might even do some Prime. I'm not being sarcastic by the way, I understand what your saying and totally agree.

If all these pass, what multiplier should I start at?

Thanks

 

[brencat]

If all these pass, what multiplier should I start at?

Thanks

200 x 15 to start

 

[Dward]

Thanks brencat, I started with 200x14.

I've run hours of OCCT, LinPack and Prime95 and all good.

When I get up to 200x17 and 200x18, if my computer BSOD's or locks up, how much Dynamic Vcore should I add exactly? Just one step?

 

[brencat]

Thanks brencat, I started with 200x14.

I've run hours of OCCT, LinPack and Prime95 and all good.

When I get up to 200x17 and 200x18, if my computer BSOD's or locks up, how much Dynamic Vcore should I add exactly? Just one step?

Haven't messed with mine since November but I think our mobos allow for +0.00625v increments which is great (but ridiculous). I believe I increased 2 steps at a time...so +0.0125v increments.

Once I got to 3.8ghz, my PC was able to pass LinX 100 passes and 3hrs OCCT High Data Set using +0.075v to dynamic vcore but Core 3 failed on Prime 95 around 9hrs. So I bumped just 1 increment at that point for a fine tune to +0.08125v and reran Prime95 and passed 14 hrs on all cores no problem.

As you can see, all tests (LinX, OCCT, and Prime95) are useful in their own way at exploiting your overclock's weaknesses if any.