3570k & Z77, not stable at 45 even with 1.376V

[tracerbullet]

3570k CPU, ASRock Extreme3 Z77 is the combo, H100i is the cooler, and in Prime 95 Blend I seem to hover around 65 deg C when things are running. Please help!


Motherboard settings that may be pertinent (* means changed from default):

Advanced Turbo 30, Optimized CPU / GPU, all = Disabled
CPU Ratio = All Core*
Multiplier = 45*
Spread Spectrum = Disabled*
Speedstep and Turbo Boost = Enabled
Additional Turbo Voltage = +.004V*
Internal PLL = Disabled*
Long Duration power Limit and all others here = 500 (Maximum)*
Power Saving Mode = Disabled
CPU Voltage = Offset*
.... -.005V*
LLC = 2*

Active Cores = All
C1E = Auto
C3, C6, Package C = Disabled*
Thermal Throttle = Enabled

***************

I started with settings from a guide most commonly referred to. I've written down some info at a few settings and what those settings were:

Multiplier (Set) / Turbo Boost (Set) / Offset (Set) / LLC (Set) / VID (Core Temp) / Vcore (CPUz) / Results
40 / +.004 / -.005 / 2 / 1.3260 / 1.304 / Ran Prime 95 1 hour, no issues
40 / +.004 / -.005 / 3 / 1.3311 / 1.312 / Ran Prime 95 1 hour, no issues (Per the guide, decided LLC = 2 was best with Vcore close to VID)
42 / +.004 / -.005 / 2 / 1.3311 / 1.312 / Ran Prime 95 1 hour, no issues
43 / +.004 / -.005 / 2 / 1.3361 / 1.328 / Ran Prime 95 1 hour, no issues
44 / +.004 / -.005 / 2 / 1.3411 / 1.320 / Ran Prime 95 1 hour, 5 x WHEA's
45 / +.004 / -.005 / 2 / BSOD (Code was 0x0000003b)

Did some testing, to see if it was all about voltage:
44 / +.004 / -.025 / 2 / 1.3411 / 1.304 / Ran Prime 95 15 minutes, WHEA's about 1 / minute (less voltage at 44 = more WHEA's)
45 / +.004 / +.030 / 2 / No Boot (More voltage at 45 than earlier but still not enough)
45 / +.004 / +.055 / 2 / 1.3411 / 1.376 / Ran Prime 95 15 minutes, WHEA's about 1 / minute (Much more voltage at 45 than earlier but still not enough)

***************

So - I'm guessing either 1) I have a BIOS setting bad (is there one that allows the use of much less voltage?), or 2) I have one of those chips that just doesn't overclock well.

All ideas / help appreciated!

 

[guskline]

What cpu cooler are you using?

 

[tracerbullet]

Ahh yes good question. H100i is the cooler, in the top of the case, pulling cool air inwards. I'm getting a pretty consistent 65 degrees C in prime 95 (give or take 5 deg). I am believing heat is not an issue.

Added that to the 1st post.

 

[Rvenger]

Enable PLL Overvoltage, turn LLC to 1.

Disable C1E, Package Cstates, C6 etc. Keep Speedstep enabled (It will throttle still with voltage drop, Overclocking with C1E and Cstates can cause the idle bug in some instances when the vcore doesn't respond fast enough to frequency)

Object is to make the Vcore more consistant at load and idle to help eliminate the WHEA log errors in event viewer.

Oh and run Prime 95 'Small FFT' test, it will surely find errors faster

 

[tracerbullet]

Enable PLL Overvoltage, turn LLC to 1... ...Disable C1E

I'll give this a whirl. I'd read about not setting LLC to 1, supposedly makes ASRock's weird out but I guess I'll find out for sure. This conflicts with Yuriman here, IIRC he has it set to 5 (essentially off).

I'll make these changes and give this a try.

 

[Rvenger]

I'll give this a whirl. I'd read about not setting LLC to 1, supposedly makes ASRock's weird out but I guess I'll find out for sure. This conflicts with Yuriman here, IIRC he has it set to 5 (essentially off).

I'll drop back to the +.004 / -.005 V settings again, make these changes and give this a try.

If I am correct, 5 will give the most Vdroop. 1 won't give any at all. If you are using the most current Bios, setting 1 should work fine.


Edit: Yep... The idea is to reduce the vdroop so your vcore doesn't drop below a certain threshold under load. I have owned a ton of Sandy and Ivy CPUs and I can tell you that each one responds to LLC differently.

 

[tracerbullet]

Yes, sorry, that's correct for this board. My UEFI looks a little different but that's essentially it, certainly so in regards to LLC.

OK, so I made the following changes per the recommendation above: Set PLL Overvolt to Enabled, LLC to 1 and C1E to disabled. Also ran Prime95 FFT instead of Blend.

Did some more testing:
45 / +.004 / +.055 / 1 / 1.3411 / 1.408 / Ran Prime 95 FFT about 5 minutes without errors, but didn't like the 1.408V and wanted to try lower settings
45 / +.004 / +.005 / 1 / Booted with a WHEA, lasted a few minutes in the desktop, then as soon as I started Prime 95 I BSOD'd (0x0_19?)

 

[Rvenger]

Yes, sorry, that's correct for this board. My UEFI looks a little different but that's essentially it, certainly so in regards to LLC.

OK, so I made the following changes per the recommendation above: Set PLL Overvolt to Enabled, LLC to 1 and C1E to disabled. Also ran Prime95 FFT instead of Blend.

Did some more testing:
45 / +.004 / +.055 / 1 / 1.3411 / 1.408 / Ran Prime 95 FFT about 5 minutes without errors, but didn't like the 1.408V and wanted to try lower settings
45 / +.004 / +.005 / 1 / Booted with a WHEA, lasted a few minutes in the desktop, then as soon as I started Prime 95 I BSOD'd (0x0_19?)

Sometimes you can increase the voltage by another +.008 and set LLC to 5. That will do some serious vdroop but it might help your low loads.

 

[tracerbullet]

Added one more data point in between the others:
45 / +.004 / +.025 / 1 / 1.3411 / 1.376 / Ran Prime 95 FFT about 1 minutes then got BSOD (0x0FC).

So w/ the PLL Overvolt, LLC to 1 (and a little less offset) and C1E to disabled - between the LLC and offset change I ended up at the same 1.376V as before, but instead of the occasional WHEA I just lost it altogether. Even at the same resulting Vcore of 1.376, this was worse, if anything, LOL.

Guess I'll try the opposite direction (lower LLC, raise offset), if one is worse the other may be better...

 

[tracerbullet]

More data points, trying LLC at the lowest setting:
42 / +.004 / +.055 / 5 / 1.3311 / 1.296 / Ran Prime95 FFT about 5 minutes, no issues (mostly wanted to see Vcore amount)
43 / +.004 / +.055 / 5 / 1.3361 / 1.296 / Ran Prime95 FFT about 5 minutes, no issues (gathering info and heading towards 45)
44 / +.004 / +.055 / 5 / 1.3411 / 1.304 / Ran Prime95 FFT about 5 minutes, WHEA's about 1 / minute (I know 45 will fail, so I'll give it a little more offset)
45 / +.004 / +.100 / 5 / 1.3411 / 1.336 / Prime95 itself failed every time I ran it, other random program failures too.
45 / +.004 / +.125 / 5 / 1.3411 / 1.376 / Boots up with a WHEA, Ran Prime95 FFT about 5 minutes, WHEA's about 1 / minute

***************

So - Even with LLC totally dropped, and Offset raised accordingly, 45 still needs that 1.376V to run at all, and does so with a lot of WHEA's.

I appreciate the help so far but it seems I've come full circle - either 1) I have a BIOS setting bad, or 2) I have one of those chips that just doesn't overclock well.

 

[BrightCandle]

Uncore/System agent voltage might also need a boost due to the increased memory interface demands.

 

[WhoBeDaPlaya]

Oh and run Prime 95 'Small FFT' test, it will surely find errors faster

Actually, I've found Intel BurnTest to be a very fast barometer of an overclock's stability.
Get pass the default 10 iterations. Then if you can pass 50-100 iterations, then throw on Prime95 (the new version with SSE) for a couple of hours.

Just noticed OP that you're using the Extreme3. I played with an Extreme4 before getting pissed of at it and swapping it for a Biostar TZ77XE4 at MC.
The LLC settings were bass ackwards (I got more Vcore fluctuations with a higher numbered setting), and IIRC, there wasn't a fixed Vcore mode, only an offset mode, which I am NOT a fan of :|

Also watch out for the VRMs. When trying to OC a 2500K, I would get throttling due to the VRMs overheating.

 

[BrightCandle]

Personally I only find LLC helps right at the extremes of overclocks, 4.5 isn't normally an extreme for todays CPUs. Why don't you try just not using LLC at all.

 

[tracerbullet]

Why don't you try just not using LLC at all.

I did - that is the #5 setting I mentioned above.

Full LLC + little offset, as well as no LLC + lots of offset, both get me around 1.376V during stress testing at 45, and in both cases I throw WHEA's all over the place.

IIRC, there wasn't a fixed Vcore mode, only an offset mode, which I am NOT a fan of

I do have a static Vcore, I'll give that a try and see if anything changes. I doubt it but will have a look.

Uncore/System agent voltage might also need a boost due to the increased memory interface demands.

I must admit I don't know enough to translate that to a variable I can adjust ...


I'm hoping there's a BIOS setting I'm overlooking. Otherwise I'll be resigned to thinking this chip needs something higher than 1.376 to hold a 45 multiplier, where it seems other people are around the 1.225-ish range. Disappointing.

 

[Kenmitch]

The bsod's you listed are not the norm that most encounter when overclocking.

What memory you using? If 2 sticks which slots?

Might have to play with other voltages it looks like.

 

[Vectronic]

He means VCCSA, I wouldn't bother touching it, my Ext4 (pretty much the same as the Ext3), steps it up from 0.925 to 1.015 or something, big jump and hasn't had any effect at all so I just leave it at 0.925.

VTT, and PLL can play a pretty big role though, I don't know if this is *supposed* to happen, but my board prefers higher VTT & PLL with increased CPU clocks, but with increased RAM clocks, it prefers lower VTT/PLL... so it's a weird balancing act.

I get 4.5GHz at 1.26 to 1.28 depending on other components (RAM and GPU), I think I'm on LLC1 right now, I'll grab a few screenshots if it will help, but I don't think you are missing any BIOS settings.

 

[Muyoso]

I had the same problem with needing ridiculous high voltage to make my 3570k stable. Sold that thing on eBay and picked up a 3770k that is so much ridiculously better.

 

[Vectronic]

Obviously an exact duplicate of these settings won't work.

Oddly, "Power Saving Mode" shouldn't be enabled, however whenever it did get enabled, hasn't seemed to cause any effect...lol

 

[tracerbullet]

The memory is this stuff, 4 sticks: http://www.newegg.com/Product/Produc...82E16820231428 (G.SKILL Ripjaws X Series, PC3 12800, for 1.5V - was free with the motherboard for Black Friday)
Should I up the voltage on it, from say 1.5 (automatically done) to perhaps 1.55? Or drop a timing or some other setting on it?

I'm up for playing w/ the other voltages. I'll accept direct recommendations of course, but don't mind doing research - any good guides on it?

I can't see any pictures right now, work is sorta stingy on what it allows through. I'll have a look tonight or soon though, do a comparison.

I appreciate the help so far!

 

[Kenmitch]

I'd up the memory voltage a little to see if it helps. Should run at stock timings I'd think. Guess you could drop the memory down to 1333MHz and see if it's the issue before upping other voltages. If memory doesn't fix you could try upping other voltages a tic or two above stock and see.

 

[Vectronic]

Your memory should be fine, 0x..3B is likely VCore, 0x...19 *could* be memory but not the memory voltage itself, probably VTT, especially if you have 4x4GB, rather than 2x4GB.

My RAM is OC'd by 400MHz, but the only reason I have the voltage set to 1.505 is so I don't have to play with it when I "Overclock" from within Windows, otherwise I can still undervolt it to 1.485 (same reason the current limits are at max, even though for 4.5GHz they can be dropped to 105)

Any way you look at it though, you're probably not going to gain anything significant, if you find the exact right voltages for VTT, PCH, PLL... you'll likely just be stable at ~1.375v... it won't magically allow you to drop it to 1.3 or something.

1.5v to 1.55v is a pretty big step, even if it was the RAM voltage, probably 1.515 would be enough to "solve" that problem.

There's really no worthwhile guides at all, anywhere, at any time.

VCCSA should be left alone, if you're going for a suicide run, and can't seem to get it stable, then try the next step up, 1.015 or whatever it is, but make sure to drop that back down to 0.925 when you're done playing.

VTT: 1.0v to 1.1v
PCH: 1.0v to 1.1v
PLL: 1.7 to 1.9v

If you can't get it stable between those limits (obviously the lower limits don't exist, 0 volts is ideal), just stop trying. If you don't really care about longevity, then add 0.15v to those.

 

[Idontcare]

The memory is this stuff, 4 sticks: http://www.newegg.com/Product/Produc...82E16820231428 (G.SKILL Ripjaws X Series, PC3 12800, for 1.5V - was free with the motherboard for Black Friday)
Should I up the voltage on it, from say 1.5 (automatically done) to perhaps 1.55? Or drop a timing or some other setting on it?

I'm up for playing w/ the other voltages. I'll accept direct recommendations of course, but don't mind doing research - any good guides on it?

I can't see any pictures right now, work is sorta stingy on what it allows through. I'll have a look tonight or soon though, do a comparison.

I appreciate the help so far!

Throw a fab over top of those and see if they respond to the better cooling. If they do then you can worry about how to make it look less ghetto, if they don't improve with the better cooling then you can pull the fan off and not worry about it.

 

[tracerbullet]

Well, still not really improved, if at all. Changed RAM to 1333 and gave it 1.510V, the olter voltages to the higher end of what Vectronic listed (all except VCCSA, didn't like the big jumps). Opened the side of the case and maxed the nearby fan speed. Didn't seem to change anything at all. Set LLC to 1 and gave it a manual 1.375 and that actually ran it with a little more voltage but still didn't make it significantly better.

I hate giving up, but I am resigned to having a chip that doesn't like to overclock. Honestly that's fine, it runs all day at 42 and that's pretty fast. Fast enough for me. It was a fun little adventure and I learned something as always.

Guys - thanks, I really appreciate the help.

 

[WhoBeDaPlaya]

Oops, fiddlesticks. Replace Extreme in my post with Pro

 

[coffeejunkee]

Well, still not really improved, if at all. Changed RAM to 1333 and gave it 1.510V, the olter voltages to the higher end of what Vectronic listed (all except VCCSA, didn't like the big jumps). Opened the side of the case and maxed the nearby fan speed. Didn't seem to change anything at all. Set LLC to 1 and gave it a manual 1.375 and that actually ran it with a little more voltage but still didn't make it significantly better.

I hate giving up, but I am resigned to having a chip that doesn't like to overclock. Honestly that's fine, it runs all day at 42 and that's pretty fast. Fast enough for me. It was a fun little adventure and I learned something as always.

Guys - thanks, I really appreciate the help.

It does look like it's just a bad chip. There was a similar thread here: http://forums.anandtech.com/showthread.php?t=2300116 and he couldn't even reach 4.3GHz. Seems like there is quite wide variety in IVB chips.

Normally, I don't think you would need to adjust any other voltage than vcore for 4.5 Ghz. Possibly disable c-states but you did that already. Maybe if you use 32GB 2133MHz memory the system agent could use a bump. But what you could try is running with 2 sticks and see if it improves in any way.