Discussion in 'Overclocking & Cooling' started by mudkip, Jan 4, 2011.
Alright this is driving me nuts. Every time I want to talk about my SB experiences I have to go sign-up on another forum. Come on, get in here peeps. Let's see your boards, chips, and settings.
P8P67 EVO + 2600k
Anything not listed is auto.
PLL Overvoltage - DISABLE
Turbo By All Cores - 46
1.365v in bios
CPU PLL Voltage: 1.7v (magic setting!)
1866 MHz, 8-9-8-24 1T
Load-Line Calibration: Ultra High
VRM Frequency: 350
Phase Control: Extreme
Duty Control: Extreme
CPU Current Capability: 130%
C1E - DISABLE
EIST - DISABLE
C3, C6 States - DISABLE
CPU Thermal Monitor - DISABLE
I had to start over from scratch a few times between crap bios updates and phantom bsods. Finally worked out 8gbs at 2133 just isn't doable with either the board or chip. Won't even do cas 7 at 1866 on sticks that will supposedly do cas 6 @ 2000mhz.
I don't have time to list all of my bios settings, but I'll give some basics. I'm running 4.6ghz on 1.365v, PLL at 1.65v. Using fixed CPU ratio instead of turbo. My chip just seems to be more stable with the fixed ratio. 8GB memory at 1866 9-10-9-27 1T. VCCIO is set at 1.05v and Vdram at 1.625v. For whatever reason, SB just seems not to like anything above 1866 for most of us.
Our chips seem pretty similar, I should give that 1.65 pll a try. On asus all you get is turbo but it can still be fixed by disabling all that crap I listed at the bottom.
I have all the CPU power saving features and other crap disabled. The deal with overclocking Sandybridge is that a stable Prime run or LinX run won't guarantee you a stable clock anymore and you will still inevitably BSOD, you'll need to keep bumping voltage until you stop experiencing crashes altogether. From a bit of googling the problem seems to be related with Vdroop on the platform itself.
My personal opinion? Overclocking just isn't as fun as you expect it to be with a 32nm chip when it is being held back by annoying repercussions such as the Vdroop, you can bench your CPU at a reasonably low voltage fine but then crash at idle leaving you no choice but to pump more Vcore into it. This is what disappointed me the most, unless it is fixed I will continue with my plan to jumpship from intel to AMD.
This is exactly what happens with SB when you are on the edge of stability, and it can be disappointing. However, considering how easy SB overclocks in general, I just look at taming the vdroop as a challenge to make up for how ridiculously easy it is to overclock otherwise. I can tell you that on newer bios revisions on my Biostar TP67XE, this problem has decreased, and voltage seems to be a bit more predictable. Major key with Sandy Bridge is to learn how to use CPU PLL and loadline calibration effectively, and you'll start seeing better stable overclocks with less vcore. Every new platform has it's challenges, and at least with Sandy Bridge it is well defined.
I did have some success in making that kind of instability show itself during stress testing. With sp1 and avx, setting IBT on standard, auto, and 100 runs. I'm not sure why, but I think the end of a round is the most likely crash point, so having it finish one every 7.5 seconds seems to bring out latent instability. Well to some extent anyways. I do have a higher setting on the LLC I can try but from what I've read it over shoots the voltage you set in the bios by a large margin.
It is fairly annoying. Watching my vcore over the span of 5 minutes I observed 1.328, 1.336, 1.344, 1.352, and 1.36. The upper range was much more common, but it only takes a moment at those lower volts to crash. I might try increasing my VRM frequency as well. Anything I can do to lock it down...
Never experienced this problem. Sounds like bad overclock testing, IMHO, crashing at idle. There are ways to avoid that, very obvious ways, actually.
Any stability testing at those settings yet? I can get mine to 4.7 on about the same vcore, and will complete 25 cycles of IBT on very high. However, like many others, the vdroop will cause random crashes on idle. If I bring the vcore up to 1.3v to make up for vdroop, it no longer BSOD randomly. Interested to see if you experience the same.
I idle @ 1600mhz. NO stability issues whatsoever. Did 500 passes of IBT with 8GB(4 sticks) @ 2133 9-10-9, stressing 7526MB, 8 threads.
That is "drooped" voltage shown above. CPU is under the IBT run in that screenshot.
I am probably running much higher current(not volts) to cpu than you guys are, FYI. I will not go over 1.3v, myself. NO need.
Memcontrol @ 1.15v.
OK. So you have Speedstep/C states enabled? What board are you using, and what sticks are you running in that board? Why do you think you're running a higher current?
Of course!?! Why would I need 4.7ghz at desktop?
Corsair CMT4GX3M2A2133C9. They need 1.7v for 2133mhz. I find 7-8-7 1866mhz faster anyway.
And that was with a very limited bios. Just use turbo to OC, and forget bios, this is like the old-school days where software is king.
There is ALOT of reason that the "real" guys are telling everyone to keep P-states enabled. I wish everyone would listen, as it's just so much easier, and saves some overall usage costs in power, too. 4.7ghz for browsing is obscene.
What board are you using? You seem to have an above average chip, probably giving you more leeway than most.
The "real" guys seem split, from what I've read over the last couple of months. I agree with you that no one needs to run 4.7 on the desktop, but there seems to be varied opinions on whether enabling C states will limit overclock stability, as it has with x58. Also, I can overclock higher using a fixed multi then I can with Turbo.
1866 seems to be the "sweet spot" for memory on Sandy Bridge.
So, what about your comment on current? Curious about that.
As do the bioses that follow them.
Bios problem. Playing with Turbo OC is no just a matter of setting volts, and running. There are various P-States that need to be programmed, and there's more than volts there to play with. Seeing boards with options out htere that are realyl basically useless kinda makes me cry.
Again, bios. But yeah, given the volts needed, 1866 seems a bit too perfect. Would be nice to see some 6-6-6 @ 1866 and 1.5v, but I'm not too confident on that at this point.
If you cannot change this in bios, or are not aware of it, then you are not using the right tools, IMHO, and your OC is really just brute-forced.
That's not to say that that is your fault, really, but more an indicator of how the bios is programmed.
Current? I am runnning 1.26v @ load, and PWM is pulling 134w. I'd be curious to see what yours is pulling.
I do not think I ahve a good chip, I'm just taking a far different approach to clocking, so our result aren't even comparable. Maybe I'd get more outta your chip than you have, maybe less...no real way to tell.
My 2600k is at 4700Ghz single core, 4600 dual core, 4500 triple and quad core. It was running 4700Ghz fine with rather toasty temperature, so I lowered it to a lower level, along with a lower vcore.
Only at 1.26v for 4.7? That is rather nice.
Actually, no, it isn't. I am doing things in a FAR different way than most, so my result is in now way comparable to yours, or really anyone else clocking SB at the moment.
Or maybe I'm just full of it. Once I get time in with some boards and more cpus, I'll have a much better picture as to what's going on.
I thought the same...boy, did I get a nice chip...but, really, I do not think so, now, after spending some time with the platform.
It's funny how we all seem to be "stuck" at the same clock though. Well, not really, but maybe you get what I mean. 4.7ghz isn't max for MY cpu, for sure.
I wonder how can you do it any different to most people? You up the turbo multiplier just like everyone else. If you are really doing it totally differently, do share it.
I am not using bios, in any way. Software only. Software programs bios, of course.
There's no trick...I'm just not adjsuting the same things you guys are. the only volts that are increased are in Pstates, cpu volts are cahnged in P-state only.
Using the latest aida the max current I saw was 128w @ 4.5 with about 1.336 loaded.
Ok, first of all, what AIDA reports, and what I am measuring is not the same number at all. For example, if you look at my review of the P7P55D-E PRO, you'll see the H55 board pulling over 90w @ stock. AIDA reports 63w.
But let's assume it's similar(given PWM efficiency, it should be, the H55 board just has crappy PWM).
You run 1.35v, and a total of 128w.
I run 1.26v, and a total 134w.
Looks like I'm pushing more to cpu, no?
ASRock P67 Pro3
4GB XMS3 @ 1600-9-9-9-24 1.65v (auto)
Hyper 212+ /w 1600 RPM S-Flex and 1500 RPM Gelid PWM (both set idle at 1000 RPM)
Turbo Multi - 45x
Voltage - Auto (idles 0.98v-1.15v, boost up to 1.37v loaded)
I can do 4.8 GHz stable with a fixed 1.42v and internal PLL overvolt enabled but what does 4.8 do that 4.5 doesn't? It seems better to use the dynamic voltage feature in the long run. Max I have been able to do is 5 GHz but it approaches thermal limits in LinX using the new AVX extension.
There's been a lot of fuss about turbo only overclocking for SB but personally I like it. It does take away from enthusiest tweaking but it also makes it very easy to achieve incredible practical overclocks while having low power consumption and heat (and therefor less noise) at idle with dynamic fan speeds on your heatsink.
OK, so let's end the mystery... What are you using in terms of software for voltage changes, current changes etc?
I was just throwing out a vaguely comparative number. If you really wanted to compare you could grab an aida beta and I could make that measurement during a max run on IBT instead of standard, and @ 4.6. Not sure any of that really tells anyone something useful though.
Also not really following your point about power states and higher work load. What board are you running and what threads have you been visiting to work out your take on it all? That would help complete the picture..
I am not visiting any threads. LoL. I've worked out my take on it all by clocking. Hence my approach not being the same as everyone else! I've read just about every thread on the internet about SB clocking, after i finished my own playing.
I own AIDA; my board does not report current consumption values...It's not even supported by AIDA at this point.
Unfortunately, the rest of the info will have to wait for my review!
Separate names with a comma.