A tale of three 920's and their OC results

yenniedn

2[H]4U
Joined
Nov 26, 2007
Messages
2,059
In the last month, I purchased three Intel i7 920 CPU's with the intention of keeping the one that OC'ed the best. My initial X58 motherboard was the Gigabyte EX58 Extreme and I had so many problems with those boards that I couldn't individually test each CPU. I RMA'ed both of the Gigabyte boards back and got an open box ASUS P6T Deluxe, which has been so much better. So without a wonky mobo to worry about, I was able to start sorting out the CPU's and the results were fairly interesting (full system specs are in my signature).

----------------------------------------------------------------------------------

i7 920 #1 (Purchased from Mwave on the week of its launch for $310 plus tax. Batch #3836A866 and pack date of 11/06/08.):

24/7 Prime95 Stable OC: 3.8GHz (20x190) @ 1.2875v. (CoreTemps) Idle temps = 36 - 40*c, Prime95 small FFT load temps = 70 - 76*c.

Highest OC tested: 4.0GHz @ >1.4250v. Idle temps = 45 - 50*c, load temps = 86 - 90*C AND NOT PRIME95 STABLE

Summary: this CPU was the earliest pack date and got to 3.6GHz with no effort. 3.8GHz was the sweet spot because BIOS vCore was relatively low at 1.2875v and idle temps were decent. At all speeds, load temps were always higher than I like to see but at 4.0GHz getting near and sometimes over 90* was very disturbing for me. The increase in voltage it needed to get from 3.8GHz to 4.0GHz was substantial (over 0.15v) with ridiculously hot temps and was definitely not worth it for me. And even then, it still wasn't Prime stable. I wasn't going to give it more voltage when temps were into the 90's as is!

----------------------------------------------------------------------------------

i7 920 #2 (Purchased from Fry's on 1/2/09 for $229.99 plus tax. Batch #3839A577 and pack date of 11/25/08.):

24/7 Prime95 Stable OC: 3.6GHz (20x180) @ 1.4000v. Idle temps = 33 - 36*c, Prime95 small FFT load temps = 63 - 67*c.

Highest OC tested: 3.8GHz (20x190) @ 1.500v!!! Prime95 hangs within 10 seconds of start. Could not boot at 4.0GHz at all even at 1.5500v!

Summary: When I took out the original 920, I went with this CPU next because it had the latest pack date and I thought that MAYBE newer is better??? WRONG!!! Since the mobo BIOS was set from original CPU above (1.2875v vCore and 20x190) I just booted this CPU at those settings. Nothing. Upped the voltage to 1.3000 then 1.3250 then 1.3500v . . . and still nothing! I thought I broke the motherboard somehow on the swap!!! When 1.4000v still wouldn't boot, I loaded defaults and it finally booted. Many trials and reboots later, I discovered that the highest BCLK this CPU would reliably boot into Windows with is 180. It took 1.325v just to boot at 3.6GHz and needed 1.4000v to become Prime stable at 3.6GHz (920 #1 only needed 1.2875v for 3.8GHz stable). It took 1.5000v just to get the CPU to boot at 3.8GHz but at this speed it was totally useless as Prime95 froze immediately each time. 4.0GHz was never reached; I stopped at 1.5500v and did not want to get into the 1.6000v range.

----------------------------------------------------------------------------------

i7 920 #3 (Purchased from Microcenter just before Christmas for $249.99 plus tax (later price adjusted to match Fry's price above). Batch #3838A819 and pack date of 11/13/08.):

24/7 Prime95 Stable OC: 4.005GHz (19x211) @ 1.4000v. Idle temps = 32 - 37*c, Prime95 small FFT load temps = 73 - 80*c.

Highest OC tested: work in progress (hope to get to at least 4.2GHz or ???)

Summary: Luckily, third time was in fact a charm! I started this CPU out at the known stable settings for CPU #1 and it booted and was Prime stable @ 3.8GHz & 1.2875v with no problems. 4.0GHz took 1.4000v but would not be Prime stable at 20x200 so I had to go 19x211 (I've read that you can do 19x211 for 4.0GHz at lower voltages than 20x200 so I'll see if I can back down the voltage a bit later on). This is the first and only 920 that I've successfully gotten to be stable at 4.0GHz and the idle/load temps are still acceptable for this speed so I'll try to push it to 4.2GHz (or more) and update if successful.

The really cool thing about this CPU is that I ALMOST returned it to Microcenter because I did pick up 920 #2 for a lower price at Fry's. But as I stood there at the return counter, I decided at the last second to keep it if the would price match Fry's price. Luckily they did and I got my keeper CPU!!! Phew!

----------------------------------------------------------------------------------

Pretty interesting how three 920's manufactured within three weeks of each other can each have different "personalities"!!! Probably most impressive is that this is done with 6x2GB Corsair Dominator DDR3 running at 1693MHz (93MHz over their rated speed). I had read that the ASUS boards had problems with 12GB but looks like BIOS 1102 might have resolved this???

Now it's time to go dump two 920's on ebay!!! :)
 
and did you mess with the uncore voltage? Heard it can make a huge difference.
 
Do you have HT and Turbo mode enabled or disabled on your overclocks??

HT IS enabled, amazingly enough. I "think" Turbo is on as well but will have to check.

and did you mess with the uncore voltage? Heard it can make a huge difference.

The disaster that was my Gigabyte EX58 Extreme experience taught me to be extremely familiar with that board's BIOS. I have only had the P6T Deluxe for about two days and haven't played with the BIOS hardly at all. I believe the only settings I adjusted are the vcore, ram voltage/timings, disabling the various power savings stuff and probably uncore voltage. This BIOS looks pretty straightforward - I just haven't had time to play with it yet. Pretty happy how easy it was to get to 4.0GHz (finally) with the little effort I put in last night.

I'll post screenshots and such later.
 
Thanks for this thread... helpful. Not boding well for my hopes to get 4.0 w/HT tho...
 
OK, I just tried 4.2GHz (20x210) at 1.4250v and 1.4500v - system would boot into Vista but would BSOD within a couple of minutes. It looks like something around 1.5000v or more will be needed to get to 4.2GHz (if possible at all). I'd do it just to see what the chip's capable of but will no doubt run it at 4.005GHz (19x211) for my 24/7 setup. I checked the BIOS and vCore was actually only 1.4000v instead of the 1.4125v I thought previously (edited above as well).

Here are the only settings I've adjusted in the BIOS:

CPU voltage = 1.4000v
CPU DLL voltage = 1.84v
QPI / DRAM Core voltage = 1.35v
DRAM BUS voltage = 1.64v
Load Line Calibration = Enabled
CPU Spread Spectrum = Disabled
PCI Spread Spectrum = Disabled
C1E = Disabled
Hyper-Threading = Enabled
RAM Timing = 8-8-8-24

Everything else on AUTO or default.

BIOS = 1102
 
What about Turbo mode? Enabled?

I'll have to check on the next reboot - running Prime95 again right now so I can't check.

Whatever the default value is on this motherboard, that's what it's set at. On the previous Gigabyte boards, default was "enabled" but I haven't seen the Core Speed jump up past the steady 4005GHz listed by CPU-Z so I'm inclined to say that it's off???
 
Since this thread doesn't mean anything without a screenshot, here's a quick one after an hour of Prime95 small FFT (temps went up a few degrees from last night due to higher ambient temps):

i79204.0primestable.jpg
 
Try lowering your UCLK to 200 and using turbo to reach 4.2ghz instead. This will relieve the stress on the memory system and should allow lower volts (on QPI for sure, should also get a few steps on vCore) so you can get the temps down a bit. I am also able to run with 8-8-8-21-1T since I'm keeping the ram at the stock speed of 1600mhz.

I heard (probably from a bad source) that you could potentially get lower volts with a lower multi and higher BCLK but I don't know how to get 211 BCLK stable on my system yet. Besides your results seem to show the opposite.

I probably have a nearly golden chip but I'm getting this so far with my i920 3841A383 from Newegg:

4.0ghz (20x200 turbo off) @ 1.40625v bios, 1.360v load with HT on, 1.375v QPI, 8-8-8-21-1T, 1.65v DRAM (1.66 BIOS), max LinX temp 83c on TRUE 120 push-pull.

4.0ghz (20x200 turbo off) @ 1.30000v bios, 1.264v load with HT off, 1.335v QPI, 8-8-8-21-1T, 1.65v DRAM (1.66 BIOS), max LinX temp 67c on TRUE 120 push-pull

3.8ghz (19x200 turbo off) @ 1.21875 bios, 1.184v load (and idle!) with HT on, 1.275v QPI, 8-8-8-21-1T, 1.60v DRAM (1.62 BIOS), max LinX temp 63c on TRUE 120 push-pull.

These are all Prime blend stable and Prime blend always requires more vCore (and vQPI for HT enabled) compared to LinX for me.

I guess you got unlucky with your Gigabyte motherboard. Maybe the BIOS you had was holding you back. The initial BIOS on the UD5 I bought was horrible but I could see how good the chip was already anyway. I initially ordered a P6T deluxe because I've been using Asus for so long but there was a supply issue so I switched to the Gigabyte UD5 because of all the good things I was reading about it on xtremesystems.

http://hardforum.com/showpost.php?p=1033562115&postcount=274
 
Nice overclocks, I would love to play with a Core i7 @ 4GHz

not to burst your bubble, but that's 4.005ghz, not 4.05ghz =D
 
Try lowering your UCLK to 200 and using turbo to reach 4.2ghz instead. This will relieve the stress on the memory system and should allow lower volts (on QPI for sure, should also get a few steps on vCore) so you can get the temps down a bit. I am also able to run with 8-8-8-21-1T since I'm keeping the ram at the stock speed of 1600mhz.

I heard (probably from a bad source) that you could potentially get lower volts with a lower multi and higher BCLK but I don't know how to get 211 BCLK stable on my system yet. Besides your results seem to show the opposite.

I probably have a nearly golden chip but I'm getting this so far with my i920 3841A383 from Newegg:

4.0ghz (20x200 turbo off) @ 1.40625v bios, 1.360v load with HT on, 1.375v QPI, 8-8-8-21-1T, 1.65v DRAM (1.66 BIOS), max LinX temp 83c on TRUE 120 push-pull.

4.0ghz (20x200 turbo off) @ 1.30000v bios, 1.264v load with HT off, 1.335v QPI, 8-8-8-21-1T, 1.65v DRAM (1.66 BIOS), max LinX temp 67c on TRUE 120 push-pull

3.8ghz (19x200 turbo off) @ 1.21875 bios, 1.184v load (and idle!) with HT on, 1.275v QPI, 8-8-8-21-1T, 1.60v DRAM (1.62 BIOS), max LinX temp 63c on TRUE 120 push-pull.

These are all Prime blend stable and Prime blend always requires more vCore (and vQPI for HT enabled) compared to LinX for me.

I guess you got unlucky with your Gigabyte motherboard. Maybe the BIOS you had was holding you back. The initial BIOS on the UD5 I bought was horrible but I could see how good the chip was already anyway. I initially ordered a P6T deluxe because I've been using Asus for so long but there was a supply issue so I switched to the Gigabyte UD5 because of all the good things I was reading about it on xtremesystems.

http://hardforum.com/showpost.php?p=1033562115&postcount=274

Thanks for the advice - I will definitely keep fine tuning it over the next couple of days. I first got to 4.0GHz by doing 20x200 @ 1.4000v but it BSOD'ed in Prime95 after about 10minutes. Instead of increasing the vcore in BIOS, I thought to lower the multi to 19x and increasing the BLCK to 211 and went back to find it's completely Prime stable. So that IS in line with what you and I have both read. Now to see if I can decrease voltage down anymore from 1.4000v or not.
 
Nice overclocks, I would love to play with a Core i7 @ 4GHz

not to burst your bubble, but that's 4.005ghz, not 4.05ghz =D

Honest mistake - not trying to cheat and get extra MHz I haven't earned ;)

Edited above; thanks for pointing it out.
 
When you first tried 20x200 @ 1.40000v bios did you have your vQPI up to 1.355 already? Try 1.375v QPI with 20x200 if you haven't yet.
 
Damn, mine sucks then. 200x20 w/ HT on
Swiftech GTZ block
Voltages:
vCore: 1.475v
vQPI: 1.52v
vCPU PLL: 1.82v
 
When you first tried 20x200 @ 1.40000v bios did you have your vQPI up to 1.355 already? Try 1.375v QPI with 20x200 if you haven't yet.

I left most of the settings consistent from one CPU to the next and only changed the vcore. QPI stayed at 1.35v all the time. I'm OK with running it at 19x211 instead of 20x200 especially if it means I can use lower voltages (which I haven't tried yet).

Damn, mine sucks then. 200x20 w/ HT on
Swiftech GTZ block
Voltages:
vCore: 1.475v
vQPI: 1.52v
vCPU PLL: 1.82v

I wouldn't say any 920 that gets to 4.0GHz really "sucks". You might need 0.075v more than the chip I have but you have watercooling to offset it so it should even out, right?
 
I guess that's true, I just have to throw a ton of voltage thru it, It'll be 3.8GHz stable at 1.225v (possibly lower) so I guess it's wall is just at the 190ish BCLK range. I may try lowering the multiplier and seeing if that helps.
 
After running at 4.4GHz for awhile, it'd be a shame to take a step back under 4.0GHz. Granted, I'm gaining 6 more physical/virtual cores, but who's counting. :p

920 @ 4.0GHz with HT enabled with a lapped TRUE Black is the goal... I'm hopin' it can happen with my "A" chip.
 
I heard (probably from a bad source) that you could potentially get lower volts with a lower multi and higher BCLK but I don't know how to get 211 BCLK stable on my system yet. Besides your results seem to show the opposite.

As it turns out, by going 19x211 instead of 20x200 (which was not stable at 1.4000v) I have been able to lower vcore by at least four notches so far. Running Prime stable right now down to 1.3750v instead of 1.4000v from earlier today.

Will continue to see how low it'll go and stay stable.
 
Well dropping the multiplier down, I'm able to do 1.400v vCore and 1.375 vQPI. Damn good improvement if I do say so myself.
 
Are you still running the same speed? I'm a big AMD fan, but if I had the cash I'd drop it on the Ci7.
 
Are you still running the same speed? I'm a big AMD fan, but if I had the cash I'd drop it on the Ci7.

Yes, I'm maintaining a 4.0GHz OC - just doing it at 19x211 instead of 20x200 (which was not stable at 1.4000v).

As it turns out, by going 19x211 instead of 20x200 (which was not stable at 1.4000v) I have been able to lower vcore by at least four notches so far. Running Prime stable right now down to 1.3750v instead of 1.4000v from earlier today.

Will continue to see how low it'll go and stay stable.

Now down to 1.3500v and still Prime95 small FFT stable!!! :eek:

That's a drop so far of 0.05v while maintaining the same speed and stability (plus dropping a couple of degrees in load temps)!!!

Put another way, that's EIGHT (8) very satisfying notches down the BIOS voltage scale!!! Woot!
 
mmm im hoping for 4ghz myself. Got a apogee GTZ coming tomorrow or the next day... got a MCR320 res and a laing pump, not sure if its mcp 350, 355 or other variant, but is one of the laing ddc pumps.
 
I had a .075v drop or 12 notch drop.
This isn't just snake oil. This is pretty neat.

LOL - I told you!!! I was just working on a screenshot for any doubters . . . but it FINALLY BSOD'ed after about 10 minutes at 1.3250v!!!

So somewhere between 1.3250 and 1.3500v is going to be my 24/7 voltage!

Screenshot will still be forthcoming! Testing 1.3375v now.
 
As it turns out, by going 19x211 instead of 20x200 (which was not stable at 1.4000v) I have been able to lower vcore by at least four notches so far. Running Prime stable right now down to 1.3750v instead of 1.4000v from earlier today.

Will continue to see how low it'll go and stay stable.

What do you have your QPI voltage set to at 19x211? And what memory multiplier are you using? Are you running your memory at 1688Mhz?
 
What do you have your QPI voltage set to at 19x211? And what memory multiplier are you using? Are you running your memory at 1688Mhz?

QPI has been constant at 1.35v.

Memory multiplier is on AUTO but is running at 1688MHz.

Turns out that 1.3500v will be as long as I can go. 1.3375v also crashed and burned on Prime. I'll get a screenie up showing 1.35v stable.
 
This thread is getting me VERY interested in the i7. Good info being passed around in here.
 
Here a quick screenie at 1.35v after about an hour:

(if you compare this screenie with the one from the first page, it looks like a 7*C drop, which is not entirely accurate. Last night I primed at about this time (colder) and the highest core was at 81*; the first screen was done midday today with higher ambient temps. At the reduced voltage, tonight's highest core is about 79* . . . so I lost about 2*C.)

i79204.0reducedvoltage2.jpg
 
I tried testing 19x211 on my Gigabyte EX58-Extreme but couldn't get it to work. It would either lock up during boot or lockup as soon as I'd start Prime95. I messed with all the voltages quite a bit but didn't seem to help. Some of the i7's may have weaker memory controllers like the AMD's so it could be that that's why I can get 4Ghz on 21x190 but not 19x211. Either that or it's my board. I'm thinking about grabbing the DFI X58 board but I don't know if I will or not.
 
Man I can't wait to build my i7 system tomorrow. Unfortunately the TRUE doesn't show up until Friday, and the sata hdd Wednesday. I burned a bootable 64-bit Ubuntu dvd to do some memtest86+ and initial stress testing and try 3.2GHz on the stock cooler! I've got this thread on my short list for voltage info. Cool stuff
 
Good job! You also have better load temps at the slightly lower voltage (vCore is 1.36v for me) than I do.
Stupid So Cal.
EDIT: I lied, my P6T overvolts a tad, 1.368v.
 
Good job! You also have better load temps at the slightly lower voltage (vCore is 1.36v for me) than I do.
Stupid So Cal.
EDIT: I lied, my P6T overvolts a tad, 1.368v.

Yeah, I noticed that the P6T does tend to overvolt. The Giga boards had really bad vdroop even with loadline calibration enabled. The 1.35v I set in the P6T BIOS shows up as 1.36v under load on CPU-Z. The same 1.35v set on my former Giga boards would've registered as something like 1.32v under load.

So it looks you and I got about the same drop. 20x200 was NOT fully stable for me at 1.40v so, let's say it would've likely taken 1.425v to get it fully stable realistically. 19x211 is fully stable for me at 1.35v or 0.075v less . . . or 12 "notches" :)

Sure am glad I didn't return this CPU!!!
 
I tried testing 19x211 on my Gigabyte EX58-Extreme but couldn't get it to work. It would either lock up during boot or lockup as soon as I'd start Prime95. I messed with all the voltages quite a bit but didn't seem to help. Some of the i7's may have weaker memory controllers like the AMD's so it could be that that's why I can get 4Ghz on 21x190 but not 19x211. Either that or it's my board. I'm thinking about grabbing the DFI X58 board but I don't know if I will or not.

I'm going to bite my tongue and not say anything too bad to trash the EX58 Extreme, but I went through two of them and am glad they are gone. If I had the same or similar problems with the P6T then I/we can look back and say maybe my issues with the Gigabyte were from operator error. Luckily, good success with the P6T gives me some vindication and a bit of reassurance that I don't need to find another hobby.
 
I actually have a new P6T Deluxe sitting here beside me that hasn't been opened. Getting ready to sell it since I already installed the EX58-Extreme.

Something I don't like about the Asus boards though is that users are reporting that Turbo mode doesn't work like it does on the Gigabyte boards and some of the other X58 boards. Asus has it set to only enable Turbo when you're under a certain TDP and when you go over that it disables. That's probably why your 21x multi wasn't working even if you had Turbo enabled in the BIOS. On my Gigabyte board the 21x multi is on all the time when I enable Turbo regardless of what the TDP or temps reach.
 
I think I've seen people get 19x211 stable on the Gigabyte boards but I don't know how they did it. I've been stable at 20x200 since BIOS F5a. I've used 19x191 with turbo which is another interesting option and stable but I'd like to dial in 19x211 with no turbo if I can!
 
Back
Top