i7 920 OC problems

[Senpuu]

Afternoon AT,

I'm trying to get my i7 920 overclocked to 3.8 on air, but am running into problems. A month or so back I messed around with OCing my CPU and was unable to continue until I had ordered some thermal grease and remounted my HS+F. At that time I was hitting 90+C at 100% load with prime95 with a very modest 160 BCLK. When I initially mounted my HS I was pretty sure I did a rather poor job of applying the grease (I think this is an extremely common problem for infrequent system builders), but after I OCed my processor slightly and stressed it, seeing the afore mentioned temps, I felt sure of it.

I have since remounted my cooler, but still feel I could have done better. I went to stock settings, autoed all the voltages in BIOS after the remount, and checked the temps: ~45C idle, ~85C loaded. I was still really unhappy with these temps, so I undervolted the CPU to 1.15, moved my rig into my living room (4-5C lower ambient), popped the case open, and tried again; the results were an improved ~38C idle and ~66C loaded. Slightly happy with the new results I tried my modest OC again without raising the voltage from 1.15 and got ~42C idle and ~70C loaded. However, I have read that Intel wants the i7s to run <= ~68C, so I don't feel terribly comfortable with running an elongated stress test at this setting to check for stability. I only ran prime95 on blend for 1hr after the OC due to being uneasy about the 70C temp.

At this point I think my best option may be to remount the HS+F again and try for a better grease job, as I think that, despite my best efforts, I have again applied too much. I honestly don't know how to use that damned applicator in such a way to get as fine a line as they want you to create on the Arctic Silver website. I'd say, just from eyeballing it, that my line was probably 1.5-2x as thick as it should have been. I've also read that some people apply grease by putting a spot in the center and then rubbing it into the CPU with a plastic-baggy covered finger, but have heard quite a few people denounce this method. Would anyone that has OCed with good temps on air like to share their experiences with that method (good or bad)? Also, would someone perhaps have a link to a good walk-through (video preferably) of someone applying thermal grease (the line and / or the finger spread method)?

I'd like to nail down whether it's my inexperienced thermal grease application(s) or an uneven HS / CPU surface that is causing my poor thermals. I figure that if I can eventually rule out an inept remounting job then I will know pretty confidently that my HS / CPU have a bad surface to surface connection. I don't really think I have it in me to lap my HS and CPU though, so I'm not quite sure where I will go from there, but I had an idea from reading some threads on AT.

I've noticed a lot of people saying that turning off HT reduces temps considerably -- sometimes seeing as much as a 10C reduction -- but I would like to keep HT on. So my idea was that I could perhaps go forward with my OCing with HT off and test for stability, assuming I see temps I am comfortable with. Then, once I have a stable OC, I could turn HT back on.

My concerns with this course of action are that the OC may not be stable after HT is reactivated and that thermals will become an issue. If an OC is stable without HT does that imply it is stable with HT, assuming there isn't a temperature-related issue? Assuming that the answer is yes, would it be a good idea to keep the OC and HT despite load temps being almost certainly ~90+C (a pretty good assumption based on what I have seen so far)? I figure that I will never see anything near 100% load in practice, so that, if stable, I will be safe from a temperature stand-point. Would this work and would it be a good idea?

Anyway, I apologize for making this so lengthy, but, if you're still reading, I'd really appreciate some help on this.

Thanks.

Rig:

CPU: i7 920 C0
Mobo: ASUS P6T Deluxe
RAM: 6GB (3x2GB) OCZ Gold 1600 C8
GPU: EVGA 260 216 (626/1350/1053) SLI
PSU: BFG R800W ES 800W
HD: WD Caviar Black 640GB
Cooler: Noctua NH-C12P

 

[imported_Shaq]

The 68C figure is measured at the heatsink and not on the CPU die which is where coretemp/realtemp measure it. 100C is max for that reading. I had mine up to 95C when testing in LinX with hyperthreading enabled, but for only a short time of course. 70C in Prime at 3.8 is about what I get at 3.9 so it should be fine. If you are OC'd to 3.8 stable at only 1.15v you have a nice chip. I need 1.4v for 3.9. You may need more voltage to make sure it is stable though. Make sure you run LinX/Linpack for 1 hour in addition to Prime (75 runs in 32bit OS or 20 runs in 64bit OS max memory). It is harder on the CPU and will raise the temps about 10C more than Prime.

You generally need a little more voltage to get hyperthreading stable at the same speed. Even though I hit 82C in LinX with hyperthreading off I rarely see temps above 60C in anything else(gaming mostly).

 

[Senpuu]

Originally posted by: Shaq
The 68C figure is measured at the heatsink and not on the CPU die which is where coretemp/realtemp measure it. 100C is max for that reading.

Well, the specs for the i7 920 C0 are as follows:

Spec Number: SLBCH
CPU Speed: 2.66 GHz
PCG: 08
Bus Speed: 4.8 GT/s
Bus/Core Ratio: 20
L3 Cache Size: 8 MB
L3 Cache Speed: 2.66 GHz
Package Type: LGA 1366
Manufacturing Technology: 45nm
Core Stepping: C0
CPUID String: 106A4h
Thermal Design Power: 130W
Thermal Specification: 67.9°C
VID Voltage Range: .80V-1.375V

Perhaps someone could define Thermal Specification for me? I was considering that the upper bound to use, but it seems I am wrong here.

Originally posted by: Shaq
I had mine up to 95C when testing in LinX with hyperthreading enabled, but for only a short time of course. 70C in Prime at 3.8 is about what I get at 3.9 so it should be fine. If you are OC'd to 3.8 stable at only 1.15v you have a nice chip. I need 1.4v for 3.9. You may need more voltage to make sure it is stable though. Make sure you run LinX/Linpack for 1 hour in addition to Prime (75 runs in 32bit OS or 20 runs in 64bit OS max memory). It is harder on the CPU and will raise the temps about 10C more than Prime.

Nope, that was at a small OC of 3.2, not 3.8. My goal is 3.8 is all I said.

Originally posted by: Shaq
You generally need a little more voltage to get hyperthreading stable at the same speed. Even though I hit 82C in LinX with hyperthreading off I rarely see temps above 60C in anything else(gaming mostly).

So then my idea won't work because I will have to play with the voltages every time I turn on HT?

 

[imported_Shaq]

If you turn hyperthreading on just lower the BCLK 4-5 notches if you don't want to raise the voltage. A .05v increase is about what you need for hyperthreading at the same frequency and that varies chip to chipand even at different frequencies. The higher the frequency the more voltage will be needed to keep it stable. Thermal Spec is different than TJmax, which is what the monitoring software uses. You will have to dig into the i7 manual on Intel's site to see it. It has the max voltages also. I have listed these below as I have read the relevant parts of the manual. The rest of the adjustable voltages don't have recommended max voltages.

100C is TJMax
1.55v CPU max
1.35v QPI/VTT max
1.88v CPU PLL max
1.85v RAM(Vddq) max Keep RAM voltage <.5v more than QPI/VTT