2500k overclocking

[keto]

OK, my finding from 2 different Z68 boards is that this chipset is VERY sensitive to memory settings. I'd be willing to place a small wager that you can get that stable, with less voltage. I suggest 1 of 2 things:

1) keep that 4.8 overclock, set your memory to 1333 speed, do your P95 and see if it still crashes. If not (I'm betting it wont), reduce voltage and repeat. THEN when you have established what your cpu will run, you go in and increase your memory speed a step at a time. Do NOT use XMP, at least not until you have tested right up to your rated speed. DO set all (or at least the 'big 5' like, 9-11-9-27-2T) memory timings MANUALLY don't let the system pull them auto. DO set your memory voltage manually.

OR

2) Just go straight to setting memory timings AND voltage manually and retry. Again, don't use XMP but you can try manually setting the timings to the same as what XMP would set.

Over and above those 2 things, other settings to mess with:
DO enable cpu pll overcurrent protection.
DO raise cpu current capability to 140% - yes, it's red. Don't worry about it.
You can also raise vccio, stock is 1.05v, I run now at 1.15v, and 1.20v is the generally accepted 'max for every day' number.

It took me a while to figure this out (from trial & error, plus the experiences of others) but I was able to drop my voltage a lot on my cpu as it wasn't the cpu that was causing the crashes, but I thought it was and kept upping vcore.

Also...as others have said, if you are seeing bench test max temps below 80C, you are fine. Thermal throttling for the cpu is 95C, so you have some headroom. Also, for *most people*, a bench test load is not indicative of what you will see in everyday usage, gaming and so forth. There's about a 15C difference for me, ie., if 70C is bench test load peak, when gaming I only see 55C.

[Patriote]

Originally Posted by Your_friendly_gamer

try using linx, just do 20 cycles!

I used IntelBurnTest 20 cycles ;) Never heard of linx?! I'll check that on google!

Originally Posted by keto

OK, my finding from 2 different Z68 boards is that this chipset is VERY sensitive to memory settings. I'd be willing to place a small wager that you can get that stable, with less voltage. I suggest 1 of 2 things:

1) keep that 4.8 overclock, set your memory to 1333 speed, do your P95 and see if it still crashes. If not (I'm betting it wont), reduce voltage and repeat. THEN when you have established what your cpu will run, you go in and increase your memory speed a step at a time. Do NOT use XMP, at least not until you have tested right up to your rated speed. DO set all (or at least the 'big 5' like, 9-11-9-27-2T) memory timings MANUALLY don't let the system pull them auto. DO set your memory voltage manually.

OR

2) Just go straight to setting memory timings AND voltage manually and retry. Again, don't use XMP but you can try manually setting the timings to the same as what XMP would set.

Over and above those 2 things, other settings to mess with:
DO enable cpu pll overcurrent protection.
DO raise cpu current capability to 140% - yes, it's red. Don't worry about it.
You can also raise vccio, stock is 1.05v, I run now at 1.15v, and 1.20v is the generally accepted 'max for every day' number.

It took me a while to figure this out (from trial & error, plus the experiences of others) but I was able to drop my voltage a lot on my cpu as it wasn't the cpu that was causing the crashes, but I thought it was and kept upping vcore.

Also...as others have said, if you are seeing bench test max temps below 80C, you are fine. Thermal throttling for the cpu is 95C, so you have some headroom. Also, for *most people*, a bench test load is not indicative of what you will see in everyday usage, gaming and so forth. There's about a 15C difference for me, ie., if 70C is bench test load peak, when gaming I only see 55C.

Thhhhaaaannnkkkssss for all the infos! I will check it all up and will post back results!

[Patriote]

Originally Posted by keto

OK, my finding from 2 different Z68 boards is that this chipset is VERY sensitive to memory settings. I'd be willing to place a small wager that you can get that stable, with less voltage. I suggest 1 of 2 things:

1) keep that 4.8 overclock, set your memory to 1333 speed, do your P95 and see if it still crashes. If not (I'm betting it wont), reduce voltage and repeat. THEN when you have established what your cpu will run, you go in and increase your memory speed a step at a time. Do NOT use XMP, at least not until you have tested right up to your rated speed. DO set all (or at least the 'big 5' like, 9-11-9-27-2T) memory timings MANUALLY don't let the system pull them auto. DO set your memory voltage manually.

OR

2) Just go straight to setting memory timings AND voltage manually and retry. Again, don't use XMP but you can try manually setting the timings to the same as what XMP would set.

Over and above those 2 things, other settings to mess with:
DO enable cpu pll overcurrent protection.
DO raise cpu current capability to 140% - yes, it's red. Don't worry about it.
You can also raise vccio, stock is 1.05v, I run now at 1.15v, and 1.20v is the generally accepted 'max for every day' number.

It took me a while to figure this out (from trial & error, plus the experiences of others) but I was able to drop my voltage a lot on my cpu as it wasn't the cpu that was causing the crashes, but I thought it was and kept upping vcore.

Also...as others have said, if you are seeing bench test max temps below 80C, you are fine. Thermal throttling for the cpu is 95C, so you have some headroom. Also, for *most people*, a bench test load is not indicative of what you will see in everyday usage, gaming and so forth. There's about a 15C difference for me, ie., if 70C is bench test load peak, when gaming I only see 55C.

Hi! Ok some updates...

I have been trying to find the lowest Vcore voltage for 4.5Ghz Prime95 Stable folowing your instructions and so far, nothing seems to be working!

Here is what I am trying out this morning.

PS: I actually just want to get a Prim95 Stable 4.5Ghz in case during summer it would get too hot running over that and that I had to drop CPU clocks as well as temps... Once I am done finding the right settings and so for 4.5Ghz Prime95 Stable, I will start to reach the Max I can while watching temps. At least I know my CPU can do 48x Multi! I actually might even hook the CPU under H20 in a near future.

I have tried setting 1.345v, 1.350v, 1.355v and 1.360v and all would make the comp post\boot but would not pass even 2 hours of Prime95 blend test even at 3.60v. Tho it latested longer at 1.360v, one of the threads would still fail giving me the error "rounded was expected at ...etc...". This morning I got mad and went ahead with 1.375v and I am now currently stress testing with Prime95 blend test.

Setting :

NOTE: I have ran 2h15 (2 Pass) of Memtest86+ at stock clocks 1600Mhz, 1.50v manual set and 9-9-9-24 2T manual set yesterday to see if my RAM had any problems which would be the reason why Prime95 blend test would fail and it returned no errors at all. Hell, I even ran 6hours of Prime95 blend test with the CPU and RAM at stock clocks to see if it would fail and It didn't! So the RAM is surely not the problem here!

http://i242.photobucket.com/albums/f...e/IMG_1743.jpg
http://i242.photobucket.com/albums/f...e/IMG_1744.jpg
http://i242.photobucket.com/albums/f...e/IMG_1745.jpg
http://i242.photobucket.com/albums/f...e/IMG_1746.jpg
http://i242.photobucket.com/albums/f...e/IMG_1747.jpg
http://i242.photobucket.com/albums/f...e/IMG_1748.jpg

Also, there is one thing I am wondering... I saw on the net that my ram was XMP 1600mhz 9-9-9-27 ... Does that means that those settings are not the stock settings?

[keto]

(Note for anyone reading the above - I'm pretty sure he means 1.360v, 1.375v, etc) Given that you are getting longer stability as you go up in vcore, it could just be that you haven't got it high enough yet....

OK
-CPU Spread Spectrum - DISABLE - this could be the big one
-C1, C3, C6 - change from auto to ENABLE (sometimes C3 and C6 can be a problem at higher overclocks and need to be disabled, but I don't see this being an issue at 4.5)
-everything else looks spot on, to me. I looked at the screenies 3x to make sure.

Question: I see what you're setting vcore at in bios, but what is it reading under load at the desktop, using a monitor like CPUZ or other hardware monitor?

You mentioned 9-9-9-27, I did see it as 9-9-9-24 in your bios. XMP is a programmed profile for your memory sticks. On some motherboards, if you select 'XMP' under memory, it will load that profile. Unfortunately, ASUS doesn't work that way - if you select XMP, it overclocks the entire system, to roughly 4.4. You can, however and as I have suggested, program those timings in manually - which you have done.

Summary - I suspect 'Spread Spectrum - Disable' might get you over the hump. However, I don't know what your actual load voltage is. I see 1.375v setting + LLC Ultra High, but I don't know what the net result of those settings is, under load, in Windows. It *should* be enough to get you stable, I think.

[Patriote]

Originally Posted by keto

(Note for anyone reading the above - I'm pretty sure he means 1.360v, 1.375v, etc) Given that you are getting longer stability as you go up in vcore, it could just be that you haven't got it high enough yet....

OK
-CPU Spread Spectrum - DISABLE - this could be the big one
-C1, C3, C6 - change from auto to ENABLE (sometimes C3 and C6 can be a problem at higher overclocks and need to be disabled, but I don't see this being an issue at 4.5)
-everything else looks spot on, to me. I looked at the screenies 3x to make sure.

Question: I see what you're setting vcore at in bios, but what is it reading under load at the desktop, using a monitor like CPUZ or other hardware monitor?

You mentioned 9-9-9-27, I did see it as 9-9-9-24 in your bios. XMP is a programmed profile for your memory sticks. On some motherboards, if you select 'XMP' under memory, it will load that profile. Unfortunately, ASUS doesn't work that way - if you select XMP, it overclocks the entire system, to roughly 4.4. You can, however and as I have suggested, program those timings in manually - which you have done.

Summary - I suspect 'Spread Spectrum - Disable' might get you over the hump. However, I don't know what your actual load voltage is. I see 1.375v setting + LLC Ultra High, but I don't know what the net result of those settings is, under load, in Windows. It *should* be enough to get you stable, I think.

Hi and thanks for the reply!

Haha yeah i meant 1.345v, 1.350v and 1.360v heh

I will try to set C1, C3, C6 to ENABLE. Currently they are on Auto. I heard that disabling those could screw the performance of my SATA 6\GBs ports which i really don't want since I am using a Crucial M4 SSD. So i am never to DISABLE them.

As for the CPU Spread Spectrum, I had not thought about it! I could aslo give that a try!

And Yeah, I also meant 9-9-9-24 2T ... Damn what's up with my typing this morning?! Maybe 3 cups of coffee was too much? I am trembling a bit... Heh Thanks for the explication about XMP!

Well, so far with the setting I posted above, I AM 2 hours of PRIME95 Blend test Stable!

With 1.375v Vcore manual set + LLC set on Ultra high, this is what i get during LOAD :

http://i242.photobucket.com/albums/f...CUltraHigh.jpg

A constent 1.360v read!

And when IDLE, i get 1.368v.

I will let it go all the way up until at least 6h of Prime95 blend test. Then, if all went fine, I will try to manually set the RAM frequency to 1600Mhz and stress test again. If all goes again fine, I'll try to disable CPU spread spectrum and see what gives. If then everything is still Stable, I might try to lower CPU Vcore now that CPU Spread Spectrum would be disabled. Do you think that CPU Spread Spectrum beeing disabled could help me get the same clocks with less CPU Vcore?

[keto]

It's possible. 1.360 is within the range of what I would expect to see for 4.5 stability, neither high nor low (probably above the mid-point, but not much).

[Patriote]

Originally Posted by keto

It's possible. 1.360 is within the range of what I would expect to see for 4.5 stability, neither high nor low (probably above the mid-point, but not much).

Cool Thanks! You are really helpful Keto!

BTW: While I'm at it, Do you happend to have any clues on how Offset voltage works? I would love to be able to get the voltage drop just like the CPU clocks when I'm IDLEing. Which is why I'd like to know more and how Offset voltage works.

Thanks!

[keto]

Yup. The chip sends a signal to the system, based on many factors but including the cpu multi, called VID or SVID. It's a voltage reading that's sort of a 'suggestion'. With offset, you don't set vcore manually, the system sets it based at least partly on the VID. The system calculated vcore can be too low, too high, or just right for your cpu. You use offset (and LLC level) to compensate for this system calculated vcore.

So, let's say you established that, setting vcore manually, you need 1.360v @ 4.5. You go into your bios, set it to 'offset mode' and, when you boot into Windows and apply a load, you see that it's only giving you 1.345. You'd then go back into bios, set offset to '+' and add .015v in the box, reboot, and you should see 1.360 at load. See below, if you change the LLC level, your offset calculation is going to change.

There is a potential problem to this - for some systems, when the idle voltage drops it's too low and can cause random BSODs. The workaround is to lower your LLC level, the lower the LLC level the higher vcore will be at idle. But lowering LLC also changes your load vcore, so you'll have to do that calculation after you change LLC and boot up, then go back and change your offset + or - (likely add more +).

[Patriote]

Originally Posted by keto

Yup. The chip sends a signal to the system, based on many factors but including the cpu multi, called VID or SVID. It's a voltage reading that's sort of a 'suggestion'. With offset, you don't set vcore manually, the system sets it based at least partly on the VID. The system calculated vcore can be too low, too high, or just right for your cpu. You use offset (and LLC level) to compensate for this system calculated vcore.

So, let's say you established that, setting vcore manually, you need 1.360v @ 4.5. You go into your bios, set it to 'offset mode' and, when you boot into Windows and apply a load, you see that it's only giving you 1.345. You'd then go back into bios, set offset to '+' and add .015v in the box, reboot, and you should see 1.360 at load. See below, if you change the LLC level, your offset calculation is going to change.

There is a potential problem to this - for some systems, when the idle voltage drops it's too low and can cause random BSODs. The workaround is to lower your LLC level, the lower the LLC level the higher vcore will be at idle. But lowering LLC also changes your load vcore, so you'll have to do that calculation after you change LLC and boot up, then go back and change your offset + or - (likely add more +).

Great! Thanks again keto!

AIDA64 says my current VID is 1.3771V. But my Vcore under LOAD won't be at 1.3711v when using Offset voltage since it will drop due to LLC right?

One question tho, when i change the setting to Offset Voltage, do I absolutely have to set either "+" or "-" along with a voltage? Because I wouldn't want to have to set a random voltage and that once I'd boot into windows have like 1.55v under load or the other way around, not be able to boot at all since the Vcore would be too low at IDLE!

Thanks!

[keto]

No, setting a + or - is optional. I don't know what your voltage will be relative to VID and LLC, you have to play with it/try it out and see. Yes, it's possible to have more vcore than you'd like, but I don't think so from where you are now. Sometimes if you're trying for a real high clock like 4.9 or 5.0+ it can happen. A short test that gave you 1.55 wouldn't hurt your cpu anyways, you have good temp control and temps are MORE important than voltage - that's why they can pump in big volts under dry ice or other advanced cooling when doing 'extreme' overclocking benchmark runs.

Oh, I've never heard of a problem booting up due to vcore being too low.