Core i7/X58 Overclocking Thread

Printable View

Show 100 post(s) from this thread on one page

03-03-2009, 06:23 PM
Kanasatake

Quote:

Originally Posted by dejanh

Indeed, I'd be extremely impressed by that. I loaded up 19x222MHz today with 1.3V but it was not stable at all. No way I could run any stress test on it. You got some nice chip there. My chip does 3.8GHz (19x200MHz) at just 1.18V, about 1.26V for 4GHz using 19x211MHz, but if I go with the 20x or higher multipliers for say 4GHz, 4.2GHz, or higher then the voltages jump way up. At 20x200MHz I need about 1.3V to be stable, or rather that was before I set up my new cooling. I can expect that I can drop that now to about 1.26-1.28 probably but I have not tried. Running 4.2 at 1.36V using 21x200MHz on the i7 940.

@Kanasatake - LinX will push your temps higher still. The TRUE does not exactly hold it's own at 4.2GHz with or without HT. Without HT you can run it ok most likely but with HT on there is no way. Not unless you can get chance's 1.30V 4.2GHz ;) I own the TRUE as well, same one as you, and it's sitting in the closet right now. I just had to ask about the temps because people get the wrong impression about what realistic clocks/temps are when reading this thread and then complain a lot about how they cannot get the same results. That's all. I did not mean to sound mean. If you do not do overly intense work that needs all 8 threads then your temps should be ok for day to day use and you should not be worries. Realistically, you got nothing to worry about until you are way up in the upper 80s or even low 90s.

I have gotten up to 3.9 Ghz with all the "goodies " on including HT, the temps were very reasonable again.

My Asus Temp alarm goes off at 65C...I just simply don't feel like turning my beautiful CPU into a hotplate. I am very happy with the results, thats good enough for me. Later on when there are better Bios updates for the new"Elipta" memory I will want to get to 2000Mhz for the memory.
I have found this chip to be the smoothest at 3.2 Ghz, with the memory @ 1600 and "all the goodies and HT on".
I do work with Photoshop and have noticed a huge performance increase when the HT is on.

Now could you explain to me the rationale of pushing your CPU temps so they can cook an egg for you? Whats in that?
As a rule I have always thought ..... 30C idle.....50C load( max)..thats been my guideline for years. What gives here, why are folks wanting to do this? Is it to kill the CPU so you can get another one?:yepp:

Lastly, so the Ultra didn't work for your temps,so the same thing happened to you as did with th folks at Hardware Canucks?....load temps in the 60C range?..Personally I though he might have gotten one of those coolers that has a bad curve on it:shrug:

Lastly I am very happy with the Rampage 2 Extreme, I am sorry there are so many folks on the Asus forum having all kinds problems. That was true when I got the Crosshair years ago, it was a "love /hate" thing.

Cheers, JRM/Hermes
03-03-2009, 06:24 PM
rge

Quote:

Originally Posted by ewitte

From what I've been reading its suggested they moved the sensor closer to the center of the core so your not actually not any hotter at 80C than you were at 60C on the other chips ;) I have no problems with 80's for stress testing as that would still be quite a bit less for everyday usage.

On 65nm and 45nm there are multiple DTS sensors on each core located in all potential hot spots, ie DTLB, L1, IntReg etc. DTS logic scans them and only reports the hottest one as core temp. See intel slide in old post below. So the i7 sensors are not "closer to the core" and even if more optimal placement occured it might be a couple C, not 20C difference. It is architectural differences, TDP, leakage, IMC, power density, etc.

On the other hand, I agree with what you said about stress testing, it is your daily computer use where temps are most important.

http://www.xtremesystems.org/Forums/...postcount=2275
03-04-2009, 03:54 AM
mojooh

Quote:

Originally Posted by SteveRo

At what voltage are you running your memory?
Memory v might need a little boost also.

I am running mem at 1.64 .... never tried 1.35 qpi though.... maybe i will give it a try when i am at the pc ...

So far i only played with vcore which didn't make a lot of difference ...

However i can see reading through the forum that not many people are running so high qpi clock and uncore even when overclocking...
03-04-2009, 04:57 AM
rge

Quote:

Originally Posted by mojooh

I am running mem at 1.64 .... never tried 1.35 qpi though.... maybe i will give it a try when i am at the pc ...

So far i only played with vcore which didn't make a lot of difference ...

However i can see reading through the forum that not many people are running so high qpi clock and uncore even when overclocking...

uncore speed = base clock x uncore multi.
qpi speed = base clock x qpi multi.

Most others are actually using higher qpi and uncore speed, but getting their by raising the base clock (bclk) as opposed to raising the multi as you are doing (since you dont want to OC the cpu).

Like Steve said, I would bump the qpi volts to 1.35, and if that works, then slowly decrease it until find stable point. Also what are your mem timings like 8,8,8 24, etc?

On the other hand, you might find it just as easy to increase bclk to get (if you change your mind about OCing cpu, then you will see increased speed, you wont see a difference by just running memory at 1333 vs 1066). Could raise bclk to 167, qpi decrease to 18, mem multi 8x, uncore to 17x, qpi to 1.35, vcore to ~1.3-1.35. That would likely be stable, then could decrease vcore (probably quite a bit) and qpi using linpack with max memory (all) as guide for stability.
03-04-2009, 05:20 AM
mojooh

thanks for the help! ....

At the moment i don't want to overclock as i enjoy all the power saving features.... when i may start rendering i would almost definetly overclock.. actually it is the first time i use a pc without overclock!

My memory is 1600mhz 7-7-7-24 @ 1.9v (now 1066 6-6-6-18 @ 1.64)
memtest at 1333mhz 7-7-7-20 @ 1.6v without problems...

I did some benchmarks with 1333mhz and 1600mhz (they don't fail .. only linpack fails...) and it does have a major impact in memory performance ...

I will try 1.35 qpi later and tell you the results ... it is possible that it could be that!
03-04-2009, 05:29 AM
rge

Yes you will see synthetic memory benchmarks like everest and sandra improve, and a small improvement on spi 32m. But on testing real world apps, games, etc very little, if any difference. If your memory is rated 1600mhz 7,7,7 24 at 1.9v that may be your problem. Did you not get i7 rated memory or is that a typo? memtest is not an accurate gauge of memory stability on i7 presumably because of the integrated mem controller. Many including myself can run memtest all day (hours and hours) without an error, but can not get linpack max memory stable on i7 using same settings without increasing mem volts.

But I would try 1.35 volts qpi first, if that fails I would try bumping up mem volts as that may be the problem. Will be interesting to see the results if your mem is rated at 1.9v.
03-04-2009, 07:07 AM
chance

Priming MULTI 21x / 4.2GHZ@1.28v (2H30) ;)

http://img14.imageshack.us/img14/9513/4221xuk8.png
03-04-2009, 07:12 AM
ReverendMaynard

Chance, use Linx at max memory please :)
03-04-2009, 07:16 AM
chance

Quote:

Originally Posted by ReverendMaynard

Chance, use Linx at max memory please :)

Don't you worry if it pass Prime95 i will pass Lynx (all memory), i already done the test several times, but when i finished Prime i wil go to Lynx ;)
03-04-2009, 07:38 AM
mojooh

Quote:

Originally Posted by rge

Yes you will see synthetic memory benchmarks like everest and sandra improve, and a small improvement on spi 32m. But on testing real world apps, games, etc very little, if any difference. If your memory is rated 1600mhz 7,7,7 24 at 1.9v that may be your problem. Did you not get i7 rated memory or is that a typo? memtest is not an accurate gauge of memory stability on i7 presumably because of the integrated mem controller. Many including myself can run memtest all day (hours and hours) without an error, but can not get linpack max memory stable on i7 using same settings without increasing mem volts.
But I would try 1.35 volts qpi first, if that fails I would try bumping up mem volts as that may be the problem. Will be interesting to see the results if your mem is rated at 1.9v.

hmmm ... didn't know that about memtest86 and i7 ! ... however the memory comes from my previous q9450 setup and there as well did not have any problems with low voltage as far as i did loosen the timings ..
I will try just out of curiosity if that is the case and will bump up the memory voltage only to test that as it could be true ... i will try 1.35 qpi and 1.75 memory...
03-04-2009, 07:53 AM
CodeNinja

I guess I need to fiddle with mine some more. Currently I'm stable at 3.8Ghz (200x19) with 1.2375v set in BIOS (shows 1.24v in CPUZ). It was running 3.8Ghz at stock voltage and -seemed- stable, but after 3 hours of Prime95 it bluescreened while working on Photoshop C4 with an "unspecified hardware error". I may drop back to stock vCore and up the QPI voltage a bit and see if that makes a difference. The processor wasn't under a heavy load, so it was probably a memory related crash.

Memory is 8,8,8,24 at 1.64v, 1600mhz, which is spec.

God I wish I wasn't stuck at work, so I could screw around with it. :(
03-04-2009, 07:59 AM
tot31

After changing my hsf from Sunbeam CCF to TRUE-120, I finally managed to get my i7 920 running at 4.2Ghz. Vcore at bios is set to 1.30625V and CPU-Z read 1.32V at 100% load.

http://img.photobucket.com/albums/v4...4hrs_prime.jpg
03-04-2009, 08:00 AM
ic3m4n2005

@chance:
You have to choose torture test with prime an use 4 threads with HT off and 8 with HT on, otherwise it proves just nothing ;)
LinX will do good aswell, best is when both of them rum stable several hours.
03-04-2009, 08:05 AM
chance

Quote:

Originally Posted by tot31

After changing my hsf from Sunbeam CCF to TRUE-120, I finally managed to get my i7 920 running at 4.2Ghz. Vcore at bios is set to 1.30625V and CPU-Z read 1.32V at 100% load.

You've got there a nice CPU too, have you try it with HTON? I need to know de diference of voltage HTON/HTOFF, im stable HTON 4.0@1.22v & 4.2@1.28v.
03-04-2009, 08:07 AM
chance

Quote:

Originally Posted by ic3m4n2005

@chance:
You have to choose torture test with prime an use 4 threads with HT off and 8 with HT on, otherwise it proves just nothing ;)
LinX will do good aswell, best is when both of them rum stable several hours.

LOL, i've used torture test (blend) and HTON / 8 threads pffff... Look P95 where is written worker#1,worker#2... #8 or CPU-Z window "8 thread". When i finished i will PS the image with the stats just for you ;)
03-04-2009, 08:12 AM
tot31

Quote:

Originally Posted by chance

You've got there a nice CPU too, have you try it with HTON? I need to know de diference of voltage HTON/HTOFF, im stable HTON 4.0@1.22v & 4.2@1.28v.

I haven't try HT On using TRUE-120 yet. I'm primming atm using ver25.9 as requested by a friend as he said that using ver25.9 the Vcore needs to be increased a bit but seems not but the max temp is raised about 1C more. Later I'll try HT On but I normally turn off HT for my daily usage.
03-04-2009, 08:14 AM
chance

Quote:

Originally Posted by tot31

I haven't try HT On using TRUE-120 yet. I'm primming atm using ver25.9 as requested by a friend as he said that using ver25.9 the Vcore needs to be increased a bit but seems not but the max temp is raised about 1C more. Later I'll try HT On but I normally turn off HT for my daily usage.

P95 v25.9, ok, ill try it later with the same settings. Thanks
03-04-2009, 08:16 AM
hlonipha

Quote:

Originally Posted by chance

Priming MULTI 21x / 4.2GHZ@1.28v (2H30) ;)

http://img14.imageshack.us/img14/9513/4221xuk8.png

one thread of p95 :rofl: :down::shocked:
hopefully this was a typo or he had the other 3 threads hidden behind the shown thread.
03-04-2009, 08:22 AM
chance

Quote:

Originally Posted by hlonipha

one thread of p95 :rofl: :down::shocked:
hopefully this was a typo or he had the other 3 threads hidden behind the shown thread.

Ok, i will give a discount for both of you. :rolleyes: Prime 95 => Window => Merge all workers. See where is written worker#1 worker#2 ... #8 at the first end of each line. #1= thread 1, #2 = thread 2... #8 = thread 8.

Other 3? Maybe you wanted to say 7, i've got 8 threads not 4. Next PS i will not merge the workers pffffff ;)

PS: Look where is "Workers - self-test, self-test, self-test..." written too. Count them :rofl:
03-04-2009, 09:46 AM
FabricioGS

Chance, seems you really have a golden chip there ;)

What's your room temp?
03-04-2009, 10:48 AM
chance

Quote:

Originally Posted by Viper.Br

Chance, seems you really have a golden chip there ;)

What's your room temp?

Yeah the chip seems to be good "enough" ;) About the room temperature i really don't know, but maybe 23/24 or even more cause its a small room (3.5Mx1.5M) and its easy to get high temperature, but the CPU it's very fresh, Noctua AIR Cooling FULL 65-70, rarely 70-75 (when priming 12> Hours) and room temperature to high.
03-04-2009, 11:31 AM
Elfear

Quote:

Originally Posted by tot31

After changing my hsf from Sunbeam CCF to TRUE-120, I finally managed to get my i7 920 running at 4.2Ghz. Vcore at bios is set to 1.30625V and CPU-Z read 1.32V at 100% load.

http://img.photobucket.com/albums/v4...4hrs_prime.jpg

What kind of difference did you see in temps between the two sinks?
03-04-2009, 03:55 PM
tot31

Quote:

Originally Posted by Elfear

What kind of difference did you see in temps between the two sinks?

These screenshots tells the difference.

Sunbeam Core Contact Freezer
http://img.photobucket.com/albums/v4...allfft_1hr.jpg

Thermalright Ultra Extreme 120
http://img.photobucket.com/albums/v4...allfft_1hr.jpg

Summary in Graph with Maximum Temperature recorded by Real Temp 3.0
http://img.photobucket.com/albums/v4...vs_true120.jpg

From the summary above running prime95 small fft, TRUE beats CCF in average of 6.25C.

CCF with LinX 20 passes
http://img.photobucket.com/albums/v4...lnx_20_ccf.jpg

TRUE 120 with LinX 20 passes
http://img.photobucket.com/albums/v4...nx_20_true.jpg

Summary in Graph with Maximum Temperature recorded by Real Temp 3.0
http://img.photobucket.com/albums/v4...ue120_linx.jpg

From the summary above running LinX, TRUE beats CCF in average of 7.5C.
03-04-2009, 05:00 PM
Ender17

what's a ballpark temperature for a i7 at 4GHz full load with decent watercooling?
03-04-2009, 05:49 PM
Elfear

Quote:

Originally Posted by tot31

From the summary above running prime95 small fft, TRUE beats CCF in average of 6.25C.

From the summary above running LinX, TRUE beats CCF in average of 7.5C.

Thanks for posting your results. I'm currently using the CCF and seeing load temps around 75C@3.9GHz 1.296V and 80C at 4.0Ghz 1.328V. Might have to consider a TRUE if I can't get these temps down.

Show 100 post(s) from this thread on one page