Originally posted by: jaredpace
Originally posted by: BonzaiDuck
First-- on the sandpaper. . . . .
10C load is probably the maximum improvement you will see - but it's worth it. More than likely lapping the IHS and HSF will give you about 4 to 6 degrees improvement on the load temps... IMO - 1 or 2 degrees minimum and 8 to 10 maximum depending on how well the connection is.
I'd agree to that without reservation. I think we're of one mind on this.
Originally posted by: Dorkenstein
Is lapping my E8600 necessary if I want to overclock it? The idea makes me nervous, since if I damage it, I'm screwed.
You're in luck, dude. My project this year started with an E8400 and ended with an E8600. I'm holding the E8400 as a replacement, or to use in another project. It was an earlier stepping in the Wolfdale line, and the news was too good for the E8600 first-issue of the E0 stepping.
There are at least a couple reasons I'd lap the sucker despite the warranty fears, and let me return to that matter just a little later here.
These Wolfdales are eminently over-clockable. In building a Wolfdale system, I wouldn't think of ending the project to run only at stock settings.
Here's the method I recommend for the lapping. There is a black plastic cover which comes with the retail-box LGA-775 CPUs, and it protects the gold-dot array of contacts. If you bought an OEM CPU, try and think of another way to do it, but the retail cover has small latches on all four edges for securing it to the CPU -- small plastic barbs. You can therefore use it to protect the otherwise-bare surfaces of greatest concern. [Intel along with general wisdom does not favor leaving fingerprints on the contacts.]
Use a surface of known flatness and strength, like a glass plate . . .[I have a 1/2" thick glass coffee-table surface]. Wet the sandpaper on both sides: it should adhere nicely to the glass. But dab off the excess water on the gritty side. Hold the plastic latches with your fingers -- securing the retail cover. Use a combination of circular motion and back and forth movement, trying to avoid having the CPU "bump" or stutter along the sandpaper surface -- it will do this because of the friction.
When you need to lubricate the paper some more, just use a couple of drops of water. My practice is to wet a paper towel and wring out a couple drops on the sandpaper as needed. Or you could use an eye-dropper. Run the sandpaper under the tap periodically to get rid of the nickel and copper particles.
You run very little -- if any -- risk of damaging the CPU if you just take a little care. Get a can of compressed air and blow off the surfaces of the processor after you finish, to get rid of any water droplets. You can dab the IHS surface periodically during the sanding with a dry, folded paper towel -- to absorb any excess moisture and metal particles.
On the risk-and-warranty issue, I really do not see any material risk, even if you lose the warranty options by lapping off the identifying information, nickel-plate, etc. The only way you'll damage the CPU is through over-clocking it.
An E8600 will clock easily with air-cooling to 4.0 Ghz. Some people are pushing them higher. But the real threat to the lifespan of the Wolfdale is the voltage and heat due to excessive over-clocking. Technically, that would void your warranty, but Intel might not detect that you "abused" the CPU, while in lapping it -- the resultant smooth copper surface -- is obvious and easy to catch for warranty "enforcement."
On my 780i motherboard, I've OC'd the E8600 to 4.1 to 4.15 Ghz with a CPU volt-setting of 1.30V. Then I backed off the speed setting to 4.0 Ghz using the same voltages for the higher setting.
The names of other voltages are different for different chipsets. You must contain your urge to increase the CPU_VTT or CPU_FSB voltage to below an actual result of 1.40V. Mine is currently set at 1.30, and I believe the sensor reading is equal to or very close to the "set" value. The 1.30V setting for the CPU is an insignificant increase over the "retail-box" maximum warranty limit of 1.25V for the Wolfdale, and in my case, it generates a sensor-reading of 1.28V at idle, and between 1.25 and 1.26V at load.
The Wolfdales -- many of them -- arrive in the retail box with defective TJunction core-temperature sensors, and Intel formally argues that the sensors were not meant to measure idle temperatures, while there is some evidence that the sensors function normally under load conditions once the temperature reaches the "stuck" value that appears at idle temperatures. So good cooling (and a colleague above has made valid criticism of the Zalman -- nevertheless sufficient) -- and lapping -- is extra "insurance" since evaluating thermal wattage and temperature using these core values is not entirely confidence-inspiring if only for the reported defect and Intel's disclaimer.
If you have monitoring software that can read the TCase sensor in addition to or instead of the TJunction core temperatures, use it to evaluate temperature at below-load settings. The TJunction temperatures will always be higher than the TCase -- from 1C to 10C greater (more specific data on the Wolfdale is available at the Intel site, but the 65nm Conroe predecessor to the Wolfdale showed a difference of about 10C according to Intel.)
I use a Noctua-NH-U12P heatpipe-tower, facilitated with a 120x38mm Panaflo exhaust fan and a 120x25mm stock Noctua fan or any other fan of the same size that is rated above about 0.15 amps @ 12V. Even the stock Intel cooler-fan is sufficient with the Noctua or a similar heatpipe cooler, if you can find a way to secure it to the cooler, and provided that an exhaust fan is situated to work as a "puller" fan to assure good air-flow.
It appears that I have too much time on my hands. I have a reputation here in the forums for "writing a book when a few sentences would do.
Good luck.