Help with 3570k overclock

makken

Golden Member
Aug 28, 2004
1,476
0
76
Hi guys. I've been playing with my 3570k clocks this weekend, and I seem to be stuck at 4.2Ghz. I can not seem to get 4.4Ghz stable at anything below 1.3vcore; and I don't want to go above 1.3vcore for longevity reasons.

My motherboard is an Asus P8Z77V-LK. Here's my BIOS settings, is there anything else I can try?

s0Ya6.png


xVjzb.png


ytRai.png


Ixr4d.png


K3TwS.png


T91s8.png


w7b12.png


Here's what I'm at right now, stable:

ymNkM.jpg


thanks
 

makken

Golden Member
Aug 28, 2004
1,476
0
76
What happens at 4.4ghz? BSOD on load?

yup, exactly.

giving it a try with PLL overvoltage enabled, and manually set at 1.250vcore...

edit: no go on 1.250 manual vcore, on 1.275 now.
I'm still noticing that its dropping to 1.208v under load, even with PLL overvoltage enabled:

BsfMb.jpg
 
Last edited:

tweakboy

Diamond Member
Jan 3, 2010
9,517
2
81
www.hammiestudios.com
That seems possible 1.30v for 4.2Ghz . It might not pass benchmarks and stress tests but as long as it doesnt crash when your working on your PC or playin games then your set. gl
 

69mustang

Member
Aug 17, 2011
81
5
71
Try adjusting your load-line calibration. I had to set mine to 16.25% on my board so that the cpu voltage wouldn't drop on load. I assume it will be different for each motherboard or power supply.
 

Idontcare

Elite Member
Oct 10, 1999
21,110
59
91
yup, exactly.

giving it a try with PLL overvoltage enabled, and manually set at 1.250vcore...

edit: no go on 1.250 manual vcore, on 1.275 now.
I'm still noticing that its dropping to 1.208v under load, even with PLL overvoltage enabled:

Your Vdroop is too much, needs more LLC.

What is your LLC set to? (it is under the Digi+ VRM menu options...it is accessible from both the BIOS as well as within AI Suite, look in the Tools section)

On my ASUS I have to set it to 75% to get best results.

Also, just so you are aware, the "CPU PLL overvoltage" doesn't reduce the Vcore needed to be stable, it actually more of a binary threshold thing. At a certain clockspeed, differs from chip to chip but is usually around 4.4-4.7GHz, the CPU won't be stable regardless of Vcore unless the "CPU PLL overvoltage" is enabled.

Enabling it will help eliminate it as a suspect in determining if you can get 4.4GHz on less than 1.3v. But your LLC is more of an issue so focus on that now as the top priority then get back to worrying about whether you need it enabled or disabled. (leave it enabled for now, disable it later on and see if you need it enabled at that point)
 

makken

Golden Member
Aug 28, 2004
1,476
0
76
Your Vdroop is too much, needs more LLC.

What is your LLC set to? (it is under the Digi+ VRM menu options...it is accessible from both the BIOS as well as within AI Suite, look in the Tools section)

On my ASUS I have to set it to 75% to get best results.

Also, just so you are aware, the "CPU PLL overvoltage" doesn't reduce the Vcore needed to be stable, it actually more of a binary threshold thing. At a certain clockspeed, differs from chip to chip but is usually around 4.4-4.7GHz, the CPU won't be stable regardless of Vcore unless the "CPU PLL overvoltage" is enabled.

Enabling it will help eliminate it as a suspect in determining if you can get 4.4GHz on less than 1.3v. But your LLC is more of an issue so focus on that now as the top priority then get back to worrying about whether you need it enabled or disabled. (leave it enabled for now, disable it later on and see if you need it enabled at that point)

Thanks. Looks good so far:
tfS9M.jpg


Seems stable at 1.23ish vcore. I currently have it manually set to 1.275vcore, I take it I should reduce my vcore 1.23ish and increase my load calibration until I don't get a drop?

How do I translate this back into offset mode?
 

Kenmitch

Diamond Member
Oct 10, 1999
8,505
2,250
136
How do I translate this back into offset mode?

Trial and error. Not sure what your MB's LLC settings are but try High/50%/? Try using offset of 0.050v's and give it a quick load to see what you get at 44x. Lower or higher the offset to get it what you need to stabalize the chip.
 

Idontcare

Elite Member
Oct 10, 1999
21,110
59
91
Thanks. Looks good so far:

Seems stable at 1.23ish vcore. I currently have it manually set to 1.275vcore, I take it I should reduce my vcore 1.23ish and increase my load calibration until I don't get a drop?

How do I translate this back into offset mode?

Note your LLC is currently set to "Medium" and your Phase Control is set to "Standard".

Bump your phase control to "Optimized" and bump up your LLC one notch higher than "Medium". (should then read "extreme" but don't get concerned by the wording which can sound bad, its not really extreme, its 75%)

Once you have those set, dial in your optimal Vcore manually.

To compute your needed "offset voltage" note the reported VID value in the CoreTemp program - be sure to only record the VID that is displayed while your computer is running fully loaded with a program like LinX/IBT/Prime95.

Also note the manual voltage you set in Asus AI Suite (do not pay attention to the Vcore reported by CPUz at this point).

The difference between the AI Suite Vcore that you manually set and the VID value you see in CoreTemp is your target "offset" value.

Right now you are seeing a 1.2410V VID and 1.275V manual Vcc, so your offset is 1.275-1.241= +0.034V

But that is with less than optimal LLC and less than optimal phase control, adjust those two first and see if that 1.275V manual setting can be further lowered.

Now you can set the offset in the BIOS, after changing your Vcore mode to offset of course, or if you are using AI Suite for your OC'ing then you need to realize that AI Suite gets a little sloppy about offset voltage labels (but it is completely manageable).

Let me know which way you intend to go for long term and we'll decide which bridge you need to worry about crossing.
 
Last edited:

makken

Golden Member
Aug 28, 2004
1,476
0
76
Note your LLC is currently set to "Medium" and your Phase Control is set to "Standard".

Bump your phase control to "Optimized" and bump up your LLC one notch higher than "Medium". (should then read "extreme" but don't get concerned by the wording which can sound bad, its not really extreme, its 75%)

Once you have those set, dial in your optimal Vcore manually.

To compute your needed "offset voltage" note the reported VID value in the CoreTemp program - be sure to only record the VID that is displayed while your computer is running fully loaded with a program like LinX/IBT/Prime95.

Also note the manual voltage you set in Asus AI Suite (do not pay attention to the Vcore reported by CPUz at this point).

The difference between the AI Suite Vcore that you manually set and the VID value you see in CoreTemp is your target "offset" value.

Right now you are seeing a 1.2410V VID and 1.275V manual Vcc, so your offset is 1.275-1.241= +0.034V

But that is with less than optimal LLC and less than optimal phase control, adjust those two first and see if that 1.275V manual setting can be further lowered.

Now you can set the offset in the BIOS, after changing your Vcore mode to offset of course, or if you are using AI Suite for your OC'ing then you need to realize that AI Suite gets a little sloppy about offset voltage labels (but it is completely manageable).

Let me know which way you intend to go for long term and we'll decide which bridge you need to worry about crossing.

Thanks for the explanation.
Here's what I ended up doing:
I changed the CPU Power Phase Control to Optimized. I'm not sure what exactly that does.
I lowered my manual vcore setting to 1.24v, then adjusted the LLC until it didn't drop below 1.23 (ended up at 'Ultra High')

I then switched back to Offset mode in BIOS, however I don't think it follows the VID exactly, because at 1.2410 VID, I should be stable, however, I BSOD on boot on anything under +0.030v

I'm at +0.050v right now. LLC at 'Ultra-High (75%)', CPU-z is fluctuating between 1.232 to 1.248 vcore @ 100% load (90% of the time its at 1.240v, with brief drops and jumps to 1.232v and 1.248v).

edit: it looks like im running out of thermal headroom at 1.240v. Core 1 is peaking at 83C
Can you actually set the offset amounts in the Asus utility? I can't find the setting for it if you can. I would like to play around with it a bit more to see if I can bring down my temps.
 
Last edited:

Idontcare

Elite Member
Oct 10, 1999
21,110
59
91
Thanks for the explanation.
Here's what I ended up doing:
I changed the CPU Power Phase Control to Optimized. I'm not sure what exactly that does.
I lowered my manual vcore setting to 1.24v, then adjusted the LLC until it didn't drop below 1.23 (ended up at 'Ultra High')

I then switched back to Offset mode in BIOS, however I don't think it follows the VID exactly, because at 1.2410 VID, I should be stable, however, I BSOD on boot on anything under +0.030v

I'm at +0.050v right now. LLC at 'Ultra-High (75%)', CPU-z is fluctuating between 1.232 to 1.248 vcore @ 100% load (90% of the time its at 1.240v, with brief drops and jumps to 1.232v and 1.248v).

edit: it looks like im running out of thermal headroom at 1.240v. Core 1 is peaking at 83C
Can you actually set the offset amounts in the Asus utility? I can't find the setting for it if you can. I would like to play around with it a bit more to see if I can bring down my temps.

Yeah the offset won't be exact. The problem with CPUz is that it quantizes the voltage readings in 0.008V increments. So you are basically flying blind to your CPU's true voltage value as applied by the BIOS as far as the 3rd decimal place is concerned, which is why you also see the offset value itself needing to be fudged ever so slightly.

83C is fine, remember the chip is OK to go to 105C but even your 83C is from a thermal stress test program. In regular usage your CPU won't even see that.

You can adjust the offset from within AI Suite, it uses the same voltage slider bar that you used to adjust the manual voltage.

This is where it gets tricky, and dangerous. AI Suite doesn't relabel the bar when you switch the bios into offset mode. So you have to realize you are not looking at the voltage or offset at that point, the voltage number reported by AI Suite is completely arbitrary. What is not arbitrary is the relative value - if it reads 1.235V when you start and you shift it to read 1.230V then you have adjusted the offset by -0.005V. If you had set the offset to 0V in the BIOS then your offset value is now -0.005V, if you had set the BIOS offset to 0.040V then your offset value is now 0.035V.

See how it is all relative? But still manageable.

Then when you dial in your offset you can go back to the BIOS and lock it in.

The danger part here comes from the fact that AI Suite will save profiles but it doesn't track whether the profile is an offset-mode profile or a manual-mode profile. I managed to jack ~2.4V into my 2600k by not realizing this D: It was brief, maybe 20s before I noticed, but not the sort of thing you want happening.
 

makken

Golden Member
Aug 28, 2004
1,476
0
76
You can adjust the offset from within AI Suite, it uses the same voltage slider bar that you used to adjust the manual voltage.

This is where it gets tricky, and dangerous. AI Suite doesn't relabel the bar when you switch the bios into offset mode. So you have to realize you are not looking at the voltage or offset at that point, the voltage number reported by AI Suite is completely arbitrary. What is not arbitrary is the relative value - if it reads 1.235V when you start and you shift it to read 1.230V then you have adjusted the offset by -0.005V. If you had set the offset to 0V in the BIOS then your offset value is now -0.005V, if you had set the BIOS offset to 0.040V then your offset value is now 0.035V.

See how it is all relative? But still manageable.

Then when you dial in your offset you can go back to the BIOS and lock it in.

The danger part here comes from the fact that AI Suite will save profiles but it doesn't track whether the profile is an offset-mode profile or a manual-mode profile. I managed to jack ~2.4V into my 2600k by not realizing this D: It was brief, maybe 20s before I noticed, but not the sort of thing you want happening.

ugh. I don't like the sound of that D:. I'll just adjust it via BIOS. I accidentally pumped 1.4v into my chip for a few seconds as well when I first started playing with the offset settings. Definitely don't want to repeat that.

Yeah the offset won't be exact. The problem with CPUz is that it quantizes the voltage readings in 0.008V increments. So you are basically flying blind to your CPU's true voltage value as applied by the BIOS as far as the 3rd decimal place is concerned, which is why you also see the offset value itself needing to be fudged ever so slightly.

Ah, makes sense. It looks like I'm at the limit of stability at +0.045 though. Another 0.005 drop and I can pass Linpack, but my display driver keeps crashing.

Seems like my chip is a bit below average; 4.4Ghz @ 1.24vcore. I believe most people were able to hit 4.4 @ 1.21vcore? anyways, hows 1.24 for long term use? I believe intel has stated that 1.3v is the unsafe level?

83C is fine, remember the chip is OK to go to 105C but even your 83C is from a thermal stress test program. In regular usage your CPU won't even see that.

This is also with an idle GPU though :p
 

Idontcare

Elite Member
Oct 10, 1999
21,110
59
91
Seems like my chip is a bit below average; 4.4Ghz @ 1.24vcore. I believe most people were able to hit 4.4 @ 1.21vcore? anyways, hows 1.24 for long term use? I believe intel has stated that 1.3v is the unsafe level?

Well don't buy into what people say is average because it is natural for people to not bother reporting their results if they think them to be merely average or below average, which means you end up getting the average value of the good chips which make it seem like the average for all chips must be really good.

If everyone who needed 1.24V with their chips actually reported their findings then you wouldn't be left thinking your chip was below average.

Also you are not delidded, correct?

My 3770k required 1.136V for 4.4GHz but that was after delidding. I didn't test 4.4GHz before delidding. But my chip was probably not average (just guessing).

As for what is safe - Intel never reported the max spec value which makes it a bit more difficult for us to determine. I think you are fine up 1.3V, maybe even higher, but until Intel tells us what Vmax is we are all just guessing.

Reminds me very much of the days before Intel would admit what TJmax was for their chips...so all the temperature programs would just guess at TJmax, needlessly so because Intel could have just published the spec on day one...same thing with these IB chips and the Vmax spec :\
 

makken

Golden Member
Aug 28, 2004
1,476
0
76
Well don't buy into what people say is average because it is natural for people to not bother reporting their results if they think them to be merely average or below average, which means you end up getting the average value of the good chips which make it seem like the average for all chips must be really good.

If everyone who needed 1.24V with their chips actually reported their findings then you wouldn't be left thinking your chip was below average.

hm, very true. Well, take my chip as one more data point for it :)

Also you are not delidded, correct?

My 3770k required 1.136V for 4.4GHz but that was after delidding. I didn't test 4.4GHz before delidding. But my chip was probably not average (just guessing).

No, not delidded. Don't intend to either, not comfortable with that level of modification for my main rig. Maybe one day when this rig is up for replacement, I'll explore that option.

As for what is safe - Intel never reported the max spec value which makes it a bit more difficult for us to determine. I think you are fine up 1.3V, maybe even higher, but until Intel tells us what Vmax is we are all just guessing.

Reminds me very much of the days before Intel would admit what TJmax was for their chips...so all the temperature programs would just guess at TJmax, needlessly so because Intel could have just published the spec on day one...same thing with these IB chips and the Vmax spec :\

Ah I remember those TJmax days well. Always hated not knowing. Until intel does release a Vmax spec, I'd prefer to remain on the conservative side of things :D

Anyways, thanks for all your help. I'm going to run some additional stress tests on both this and the GPU and then decide if I want to keep it at 4.4 or drop back down to 4.2.
 

Eureka

Diamond Member
Sep 6, 2005
3,822
1
81
Just wondering, are you using an aftermarket cooler? 83C at 1.24V seems a bit hot.
 

Eureka

Diamond Member
Sep 6, 2005
3,822
1
81
What's your ambient temperature? Mine usually peaks around 60-70C with a CAFA50 with my i5-3570k at 1.3V.
 

makken

Golden Member
Aug 28, 2004
1,476
0
76
What's your ambient temperature? Mine usually peaks around 60-70C with a CAFA50 with my i5-3570k at 1.3V.

Room temp hoovers around 18 to 20C here. I also have a second 120mm fan attached in a push/pull config

I'm going to try to reseat the HSF this weekend, and probably add more TIM. that Coolermaster TIM was really thick and didn't spread all that well honestly.

may try http://www.newegg.com/Product/Produc...82E16835100017 to see if it helps...
 

Idontcare

Elite Member
Oct 10, 1999
21,110
59
91
Room temp hoovers around 18 to 20C here. I also have a second 120mm fan attached in a push/pull config

I'm going to try to reseat the HSF this weekend, and probably add more TIM. that Coolermaster TIM was really thick and didn't spread all that well honestly.

may try http://www.newegg.com/Product/Produc...82E16835100017 to see if it helps...

Ceramique is good stuff, but if you are going to go to the trouble of ordering new TIM and remounting your HSF then you may as well go with NT-H1, $10 shipped.

http://forums.anandtech.com/showthread.php?t=2285595

MasterChartforbare-dieTIMtesting.png
 

makken

Golden Member
Aug 28, 2004
1,476
0
76
Ceramique is good stuff, but if you are going to go to the trouble of ordering new TIM and remounting your HSF then you may as well go with NT-H1, $10 shipped.

http://forums.anandtech.com/showthread.php?t=2285595

MasterChartforbare-dieTIMtesting.png

Yea I choose ceramique based on your results in that thread. It looks quite good considering you didn't even bother with burn in. Your comments on ease of use of it pushed me in favor of it over NT-H1, honestly, as I tend to be pretty bad about applying that stuff. (Maybe I should stop using old credit cards)


Either way I already have an order placed for it.
 

Idontcare

Elite Member
Oct 10, 1999
21,110
59
91
Yea I choose ceramique based on your results in that thread. It looks quite good considering you didn't even bother with burn in. Your comments on ease of use of it pushed me in favor of it over NT-H1, honestly, as I tend to be pretty bad about applying that stuff. (Maybe I should stop using old credit cards)


Either way I already have an order placed for it.

Well I can't argue with you when you've gone and deftly used my own words against me :p

It is true, ceramique is silly good at getting good mounts. You will not be displeased.
 

makken

Golden Member
Aug 28, 2004
1,476
0
76
Remounted the HSF this weekend, it looks like load temps dropped about 4 degrees to 78C. With my room hovering at 18C, that's 60C over ambient.

Anandtech's review of my case using my HSF got 58.9C over ambient, so I'm about in line with those results. Granted, they are overvolting to 1.38V, and I'm only at 1.24V.

Short of delidding, don't know if there's anything else I can do to get the temps down. I was hoping to get it low enough to see if I can hit 4.5Ghz under 1.3V, but this is still a bit too hot for me to push any further comfortably.
 

2is

Diamond Member
Apr 8, 2012
4,281
131
106
I personally would enable internal PLL overvoltage before using EXTREME levels of LLC. In fact, if I have to set LLC that high, it's not worth running the CPU at that speed, as you're clearly trying way too hard to force the CPU to do a speed it really doesn't want to do. I've also noticed that on Ivy Bridge, Internal PLL Overvoltage yields benefits at much lower clock speeds than previous CPU's.