Core i7/X58 Overclocking Thread

[safan80]

When you guys post your OC can you guys state which bios your using?

so many different bioses for a lot of the boards.

[Hot'n'Chilly]

I'm on the newest P6T bios whichever that is... just flashed it about a week ago.

Anyway, do you guys know which voltages you can drop besides CPU voltage to effect overall heat? Does dropping the DRAM, PLL, QPI (and others) effect temps?

[Xello]

They're all voltages that are going into the cpu itself so they will affect it to some degree, which voltages will make a difference to your cores though (the temps you care about) are up for question.

You should try to run with lowest volts possible anyway, on all settings

[Hot'n'Chilly]

Looks like I have some detective work to do then. I'll post in if I find anything interesting. And anyone else with info please chime in.

[mremulator]

I'm not sure if you can trust that or not, i passed 50 runs yesterday on linX (2hr 50min~) then proceeded to crash after 9hr of blend. I know it's 3hr vs 9hr but from what pretty much everyone says 10 mins of linpack is supposed to be more reliable than 10 hours of blend. I don't know why but i had a feeling this would happen, i'll probably be sticking with prime after all.

Have a look where prime crashes on blend... mine kept crashing on 4096 FFTs, however, ran 24/7 with small FFTs only. Therefore, I concluded that it was down to memory timings/voltage and not the CPU. So I raised the Vdimm a little and slackened off my RAM timings... it still crashed?? So... I lowered the BCLK from 200 to 190, now it's stable @ 4.0GHz (21x190) and it could only do 3.8 before. Yea I know it's still 3.8GHz for multi-threaded apps, however, I have an "A" revision chip and I'm happy with 3.8GHz stable for 24/7 use.

Try running a "custom" session of Prime with min/max FFT size set to 4096 for an hour and see what happens. The way I understand it is small FFTs for CPU testing and large FFTs for memory. So I run small FFTs (12 hours+), blend (12 hours+) and custom 4096 FFTs (1 hour). Then I consider it stable.

[Xello]

Have a look where prime crashes on blend... mine kept crashing on 4096 FFTs, however, ran 24/7 with small FFTs only. Therefore, I concluded that it was down to memory timings/voltage and not the CPU. So I raised the Vdimm a little and slackened off my RAM timings... it still crashed?? So... I lowered the BCLK from 200 to 190, now it's stable @ 4.0GHz (21x190) and it could only do 3.8 before. Yea I know it's still 3.8GHz for multi-threaded apps, however, I have an "A" revision chip and I'm happy with 3.8GHz stable for 24/7 use.

Try running a "custom" session of Prime with min/max FFT size set to 4096 for an hour and see what happens. The way I understand it is small FFTs for CPU testing and large FFTs for memory. So I run small FFTs (12 hours+), blend (12 hours+) and custom 4096 FFTs (1 hour). Then I consider it stable.

Good point, i might look into that later. I wonder if you can set 5000k+ FFT's, to use up the same amount of memory linpack does.

Don't pay much attention to a/b batch [note: not revision!], too early in the manufacturing process for it to be that clear cut i think. I have an A and can go 4.2ghz+4.3ghz stable, and it's also one of the coolest chips i've seen, topping 60c~ @ 1.45-1.47v vcore

[dejanh]

Wanna trade? :p

[mremulator]

I have an A and can go 4.2ghz+4.3ghz stable, and it's also one of the coolest chips i've seen, topping 60c~ @ 1.45-1.47v vcore

HT off I assume?

[Xello]

HT off I assume?

Yeah, not sure how much temps increase with it on, i'm told 10-15c but i might find out for myself one rainy day.

[Hot'n'Chilly]

I have a request. Will those of you with stable 3.8ghz & 4.0ghz OC's on air please post in with your current settings and Voltages? There have not been very many people posting in with full reports in the past few weeks, so I thought it was about time for some new ones.

Thx

[SteveRo]

my 4.0 on air - this is also a doable do with cpu v at 1.4v, qpi at 1.34v and dram at 1.64v

[Tobit]

So, my i7 920 will be here on Monday or Tuesday. I will be initially running on air using a Noctua NH-U12P SE1366 on the bench while I decide on what case I want to put this new build in. I was going to order a TRUE but was having a hard time finding a vendor who had the 1366 bracket in stock. I will be going water with a dedicated CPU loop ultimately. Anyway, in my tool box, I have both AS5 and Ceramique. Any idea which one I should try first with the 920? I suppose I could try both, just trying to see if anyone has already been down this road. Thanks.

[Splave]

AS5 unless you are going sub zero

[Hot'n'Chilly]

my 4.0 on air - this is also a doable do with cpu v at 1.4v, qpi at 1.34v and dram at 1.64v

Interesting, I'm not sure my CPU would be stable at those settings though (its picky) but the 30x/134 setting is very interesting. What made you come to this setting... is it more stable than having a lower multi and higher BCLK?

[DavidNJ]

A hypothesis is as bclk passes 200Mhz the QPI becomes problematic, the exact cut off depending on the chip, mobo (probably the quality of the buck converter), and maybe temps (just guessing that 90C is less stable than 50C, at least for a few Mhz).

The 965 lets you get past that, but there are SO many 200x21s it seems that is the Q6600 G0 equivalent of 3.6GHz. The question is what is the limit in the 4.4-4.6GHz range? Is it QPI, which means that the 940 may show an edge of the 920? Or is it the CPU itself, which would mean the 920, 940, and 965 are functionally equivalent?

[Hot'n'Chilly]

So would you suggest that those of us who have troublw at 20x/200 try something more along the lines of what SteveRo did, with a 30x/134? Is there any danger in using unbalanced combos such as that? Or is it in fact more dangerous to drive the BCLK up (like 200+)?

[DavidNJ]

If you have a 965 you can try everything independently and use the Bclk for granularity between multipliers. If you can't get a 200 Bclk I would think that tuning that independently would be the first step. The QPI is 3.6GHz which is fairly close to the stock 3.2Ghz 965 setting. Can you make 15x200? It should be similar to any other system. I tried, so far unsuccessfully, to get posts of Core i7 settings in a different thread. Otherwise, they are spread all over.

Have you seen the NCIX video or read Saaya's thread?

[Hot'n'Chilly]

Yes, I've seen the video. I was simply wondering about the merits of high multiplyer with low BCLK vs those of low multiplyer with high BCLK.

By the way, I believe I have reached a Prime stable 3.8ghz overclock. It just passed through 4 hours of torture testing with all 8 threads, with a top max of 77c and an average max of 75c.

It is as follows:

CPU Ratio: 19x
BCLK: 200
DRAM: 1203mhz (note to non P6T users: the P6T does frequency rather than multi's)
UCLK: 3208mhz
QPI: 7218MTs
-----------------
CPU Voltage: 1.32v
PLL: 1.84v
QPI Voltage: 1.3375v
DRAM Voltage: 1.6v



This was done with an air cooled (Noctua) 920 on a P6T board using OCZ Platinum DDR3 1333 RAM.

[Chicken Patty]

Yes, I've seen the video. I was simply wondering about the merits of high multiplyer with low BCLK vs those of low multiplyer with high BCLK.

By the way, I believe I have reached a Prime stable 3.8ghz overclock. It just passed through 4 hours of torture testing with all 8 threads, with a top max of 77c and an average max of 75c.

It is as follows:

CPU Ratio: 19x
BCLK: 200
DRAM: 1203mhz (note to non P6T users: the P6T does frequency rather than multi's)
UCLK: 3208mhz
QPI: 7218MTs
-----------------
CPU Voltage: 1.32v
PLL: 1.84v
QPI Voltage: 1.3375v
DRAM Voltage: 1.6v


This was done with an air cooled (Noctua) 920 on a P6T board using OCZ Platinum DDR3 1333 RAM.

similar setup to mine now

Here is my daily setup for now, still going higher day by day:

Got a shot of my background so you'll can see what the rig looks like as well

quick run in wprime so you can have an idea of what temps are like, dont get any hotter over the hours

[SteveRo]

Interesting, I'm not sure my CPU would be stable at those settings though (its picky) but the 30x/134 setting is very interesting. What made you come to this setting... is it more stable than having a lower multi and higher BCLK?

I have found (so far) - for HT on - I have better success with high multi, low bclock. Just the reverse for HT off - lower v's and temps come with low multi and high bclock -

[SteveRo]

A hypothesis is as bclk passes 200Mhz the QPI becomes problematic, the exact cut off depending on the chip, mobo (probably the quality of the buck converter), and maybe temps (just guessing that 90C is less stable than 50C, at least for a few Mhz).

The 965 lets you get past that, but there are SO many 200x21s it seems that is the Q6600 G0 equivalent of 3.6GHz. The question is what is the limit in the 4.4-4.6GHz range? Is it QPI, which means that the 940 may show an edge of the 920? Or is it the CPU itself, which would mean the 920, 940, and 965 are functionally equivalent?

David, based on what I have seen so far (previous post) and comparing to what others have come up with - I think 920, 940 and 965 are all the same (within manufacturing tolerances) with the only exception being the 965 multi unlock.

[safan80]

David, based on what I have seen so far (previous post) and comparing to what others have come up with - I think 920, 940 and 965 are all the same (within manufacturing tolerances) with the only exception being the 965 multi unlock.

I agree... the 965 does not seem any better using the same multis as the 920 or 940.

[SteveRo]

David, based on what I have seen so far (previous post) and comparing to what others have come up with - I think 920, 940 and 965 are all the same (within manufacturing tolerances) with the only exception being the 965 multi unlock.

One other exception - A vs. B batch look like might make a big difference in oc. Maybe more tolerant IMC? I think most folks that are getting 4.2/HT on stable are probably B batch. Mine is an A

[SteveRo]

How many cores does crysis use?
Can we get more frame rate if we oc just 1 or 2 cores of an i7? (turn off unused cores?)

[Chicken Patty]

One other exception - A vs. B batch look like might make a big difference in oc. Maybe more tolerant IMC? I think most folks that are getting 4.2/HT on stable are probably B batch. Mine is an A

i ran at 4.2 GHz stable for about one week until I decided to re overclock the whole thing, i have it saved in my bios though, I can just go and load it, works like a charm. HT Enabled of course, if you want some benches let me know, I can post them