i7-920 Unable to oc over 3850MHz

A

axdx

Limp Gawd
Joined
Sep 25, 2003
Messages
242
As the topic says, I seem to be completely unable to get past 3850MHz mark
Have tried up to vcore 1.32V, system seems stable at 3840MHz, but as soon as i hit 3860MHz i start getting prime95 errors and BSoD

System seems stable so far with the following settings:

BCK = 213
CPU speed = 3841MHz
CPU multiplyer = 18
Memory speed = 1706MHz @ 7-8-7-20-2T
vcore = 1.268V
pll = 1.88V
qpi = 1.35V
DRAM bus = 1.65V

Temperatures are a bit high though, up to 82C during torture tests.

Any ideas?
 
Last edited:
I run mine at 1.35 voltage but you might want to remove your ram as the limiter...try dropping the ddr speed and find the highest overclock you can get stable just from your cpu, also you don't mention what cooling you use 80's is not that high for a i7 but high 80s to 90s will cause you to bsod.
 
I have tried running BCK 191 with multiplyer @ 21 and memory at the lowest speed. It wasnt stable at 1.32V, i think it might have been stable at 1.35V but temperatures were extreme (up to 91C) so i didnt test that long

I am using a zalman air cooler, and its probably worth mentioning i got no intake or exhaust fans on this case due to stuff not fitting. Can't wait for the corsair case to arrive...
 
I wouldn't attempt any further OCs when you hit 90s... I would personally stop at 70C in LinX.
 
I wouldn't attempt any further OCs when you hit 90s...
Click to expand...

Agreed. Also, with no intake or exhaust the air's only going to keep recirculating and getting hotter...

Additionally, BCLK of 213 is pretty high for any kind of a standard voltage - hoping to overclock past where you're at you'll most likely be looking at dropping the BCLK and upping your CPU multi.
 
running 200 bclk with 21 multi = 4.2 ghz stable d0 920

vcore is 1.33
qpi - 1.35

all other voltages set manually to standard setting.


memory is running at 1600 stock.
 
running 200 bclk with 21 multi = 4.2 ghz stable d0 920

vcore is 1.33
qpi - 1.35

all other voltages set manually to standard setting.


memory is running at 1600 stock.
Click to expand...

That's nice but how does it help OP?

axdx, when that corsair case comes in and you have some better cooling try the 1.35v setting. That could very well be your problem. It could also be your chip just can't do 4ghz.
 
Sorry I meant to post this as well earlier.

http://whirlpool.net.au/wiki/?tag=rmp_i7_920_overclocking

might want to stop and look at this.... your bclk is very high - if you're going for cpu clock you probably need to raise your multi to 19 or even 21 with turbo enabled. ram is overclocked if is it 1600 so that could cause some instability as well. I just think taking components out of the equation is ideal to really fix the issue such as the cpu core or other areas receiving enough voltage.

Very good guide. Followed it verbatim and ran 3.66 at 1.1v - stock on everything else.
 
noteworthy said:
Sorry I meant to post this as well earlier.

http://whirlpool.net.au/wiki/?tag=rmp_i7_920_overclocking

might want to stop and look at this.... your bclk is very high - if you're going for cpu clock you probably need to raise your multi to 19 or even 21 with turbo enabled. ram is overclocked if is it 1600 so that could cause some instability as well. I just think taking components out of the equation is ideal to really fix the issue such as the cpu core or other areas receiving enough voltage.

Very good guide. Followed it verbatim and ran 3.66 at 1.1v - stock on everything else.
Click to expand...

Thats what I started with. 210 BCLK and 19 multi (vcore 1.32). Memory was at whatever the lowest settings is. I did it pretty much just to check if CPU could handle it before I start on RAM - it didnt. System ran for a while before throwing a BSOD or prime95 error.

After that I switched to 191 BCLK 21 multi - it didnt pass the torture tests either. Then I spent a few hours testing configurations with 190 - 205 BCLK 19/21multi, but failed to find any stable settings.
Bumping vcore up a little didnt seem to help, so I set 19 multi and started decreasing BCLK until I hit stable (which was just under 3850MHz). At that point i was still running 1.32 vcore, so i started dropping it until I hit minimum stable at that speed (1.27). After that I figured I might as well get all the bandwidth out of the RAM - and thats how I ended up with my current settings. Strikes me as kinda odd that it would run 3848MHz at 1.27V, but at 3865MHz even 1.32 is not enough.
 
Last edited:
Bump your QPI PLL Vcore by a little.
Bump your PCI-E frequency up a tad (if you are not running an SSD)
Bump your VTT (in your motherboard, also known as QPI).

If your high BCLK is really stable at 213, that means it should hit 213 * multiplier with the right the settings.

I suggest reading more about OCing before continuing, especially if you are hitting 90Cs and not sure what your BSODs mean. If your BSOD is a 0x124 or 0x9c, CPU VCore does not help, you need CPU VTT (Your motherboard QPI voltage).
 
noteworthy said:
running 200 bclk with 21 multi = 4.2 ghz stable d0 920

vcore is 1.33
qpi - 1.35

all other voltages set manually to standard setting.


memory is running at 1600 stock.
Click to expand...

Each chip overclocks differently...So a 3.8GHz OC isn't always guaranteed.
 
MJZ said:
Bump your QPI PLL Vcore by a little.
Bump your PCI-E frequency up a tad (if you are not running an SSD)
Bump your VTT (in your motherboard, also known as QPI).

If your high BCLK is really stable at 213, that means it should hit 213 * multiplier with the right the settings.

I suggest reading more about OCing before continuing, especially if you are hitting 90Cs and not sure what your BSODs mean. If your BSOD is a 0x124 or 0x9c, CPU VCore does not help, you need CPU VTT (Your motherboard QPI voltage).
Click to expand...

I've done all that apart form PCI-E freq. Booting with CPU speed over 4ghz caused a 0x124 BSoD within max 2 minutes. Bumping QPI up a couple notches fixed that. However both lynx and prime95 still errored out after just a couple tests, and there were occasional random application crashes, which is a pretty sure sign of CPU making errors.
 
I went back to trying to get the CPU to run at 4ghz, and I just noticed that some of my voltage settings are misbehaving. Badly.

Load line calibration is evil. Looks like when CPU is under high load, if LLC is enabled it will make voltage increase instead of decrease. Not a little either, acutally increased by 0.05. No wonder when i tried with 1.35V (aka 1.4V at full load, thanx LLC) my temps were almost 90.

LLC is disabled forever now. CPU is stable at 4.1MHz, i think its pretty much as much as it will OC @ 1.35V. So, now the question is - should i go for 210ish BCLK and 19 multi to run RAM faster (BCLK 195 multi 21 atm)?
 
Last edited:
zero940 said:
Yours is the dance of death! Seriously though, do you have the money to replace the i7 920 if it were to brick on you tomorrow?
Click to expand...
Modern CPUs will turn themselves off before they run hot enough to damage themselves. I've had several CPUs shut themselves down due to either insufficient cooling or the the lack of water flowing in my loop. All of the CPUs that I've had this occur with went on to live normal lives.
 
TeeJay said:
Each chip overclocks differently...So a 3.8GHz OC isn't always guaranteed.
Click to expand...


+1. Every chip is different. And unless it is an extreme edition or a black edition then the CPU isn't "supposed" to overclock alot You have a 1ghz gain I would suggest taking what you have and running with it

edit:your sig says your at 4.1 is that a typo ??
 
that high a heat will still significantly reduce the cpu's life span, whether it's considered 'non-damaging' by the cpu it's self or not. it's pretty much the difference between using a chainsaw or a hand saw (at least from my understanding, something like every 10C or so over what's normally considered 'good/safe' cuts ~2/5 the expected life off).
 
Last edited:
W.Feather said:
+1. Every chip is different. And unless it is an extreme edition or a black edition then the CPU isn't "supposed" to overclock alot You have a 1ghz gain I would suggest taking what you have and running with it

edit:your sig says your at 4.1 is that a typo ??
Click to expand...

No, he broke the 3.8ghz barrier (with my helpful tips ;) )
 
Leon_Solitario said:
that high a heat will still significantly reduce the cpu's life span, whether it's considered 'non-damaging' by the cpu it's self or not. it's pretty much the difference between using a chainsaw or a hand saw (at least from my understanding, something like every 10C or so over what's normally considered 'good/safe' cuts ~2/5 the expected life off).
Click to expand...
Did you even calculate out what it would mean if your "facts" were correct? If a CPU normally has a 30 year life it would essentially be dead after running 10C over its safety span three times (it would lose 12 years each time). I'm not saying extreme temps are good for a CPU and it will not impact its life, but I will say that unless you have substantial (multiple controlled studies) evidence to back up your claim, I'm going to call bullshit on your facts.
 
I want to say it was from an OC guide I read on Tom's hardware, but could be wrong, I've been doing too much reading the past two weeks to recall, lol. the source didn't have any hard evidence to back it up other than a bit of math on electromigration. this was done for 'older' intel (~3+ years) cpus, although I'm not sure that materials have been improved enough to totally throw off the estimations.

I'm sure the equations are probably out there to get more accurate values, but heat does definitely have an exponentially growing detrimental effect on any hardware (the affect of which being different depending on the hardware obviously). so much so that I wouldn't discount such a claim at least, though it does seem a bit extreme (I would argue myself that it's exaggerated).

just to clarify, the "safe/good' temp I was referring to would be manufacturer's specs. life expectancy seems to be 10+ years for newer processors running stock, so even running a more extreme temp you wouldn't normally see the processor die before you changed it out/upgraded.
 
I got the system running stable now with the specs in my sig. I had quite a few problems getting there though.

The biggest problem was vcore. As far as i can tell, my CPU needs a minimum 1.32V to be stable at 4.2GHz. Due to actual voltage being lower then what is set in BIOS and voltage drop at high load, I would have had to set 1.38-1.39V vcore in BIOS without use of LLC (1.35 actual voltage at no load, 1.32 at high load). I chose instead to re-enable LLC and set vcore to 1.35. With those setting CPU runs at 1.32V at no load and 1.35 at high load.

The other problem was memory. For some reason the board becomes unstable at speeds over 1600. When I ran at 200 BCLK and 1600+ memory (at C7 timings) I got BSoD x124. Bumping QPI up fixed that particular error, however during small FFT tests I would instead get "non paged pool" or "irq not equal or less" BSoDs. Further increas of QPI did not help. Instead I chose to back down to 197BCLK with 1580MHz memory speed and tighten the timings to 6-7-6-18. The memory is rated at 1866 7-8-7-20.
Seems like the board just cant keep up at high memory speed. I did try loading default BIOS settings and using the memory profile to run it at 1866MHz with correct timings and all, but the system was not stable. ("non paged" and "irq blah blah" BSoDs)
 
axdx said:
As the topic says, I seem to be completely unable to get past 3850MHz mark
Have tried up to vcore 1.32V, system seems stable at 3840MHz, but as soon as i hit 3860MHz i start getting prime95 errors and BSoD

System seems stable so far with the following settings:

BCK = 213
CPU speed = 3841MHz
CPU multiplyer = 18
Memory speed = 1706MHz @ 7-8-7-20-2T
vcore = 1.268V
pll = 1.88V
qpi = 1.35V
DRAM bus = 1.65V

Temperatures are a bit high though, up to 82C during torture tests.

Any ideas?
Click to expand...

If you ask me you're running way too high a BCLK, why using only the 18x multiplier?

For 4Ghz stable 191 x 21 is by far the most stable setting providing your motherboard supports using the turbo multiplier, it will also require way lower voltages.

Also you're really pushing the limit of the memory controller at 1700Mhz.
 
Yea, it isn't your board that can't run the memory at 1866, it is your CPU. It depends on how well you can overclock your Uncore.
 
i know you are afraid of bumping up the voltages.

try 1.5 qpi and 1.4 cpu with everything else on AUTO. move qpi down little by little until it's stable then not stable again, then nove it back up one notch. then move cpu down little by little until not stable, move it back up. There are 2 codes you should be looking for (1 for pqi and 1 for cpu voltages).

at one point, i had mine at 1.515 and 1.4. Then optimized them over and over again to hit the sweet spot.

I got mine to 4.2Ghz with HT turned off, that HT was burning up my chip!! ;)
also, like others are saying, why not 21x?
 
now my qpi and cpu are 1.4something and 1.3 (read 1.26 in CPUz) respectively.
 
You must log in or register to reply here.
Back
Top