Subject: Essay (honestly) on Intel CPU thermals
By: Ray (IP: 81.77.54.*)
Written on: 27-11-2006 03:30
Background:- I had a spare few hours and a nearly-dead Intel socket 478 2.4C CPU left over from an incident with a motherboard that decided to switch off all the PC's fans - including the processor fan - whilst still continuing to run the burn-in test that I'd started. Not Good ... the built-in thermal protection & fan monitors ignored the situation completely and all I got was an alarm from the motherboard buzzer. It didn't help that I wasn't near the PC at the time, the CPU reached 100C and after weeks of intermittent faults the CPU & motherboard were retired.
I'd read about how demon overclockers lap the top of their Intel IHS to lower temperatures. So I decided to give it a try for myself, just to see what was involved, how hard it would be and what lessons I could carry forward. I also suspected that the original thermal compound under the IHS would be in pretty poor shape.
a) IHS removed. My badly worked removal caused deep scratch marks in the green circuit board. Lesson - don't cut down into the circuit board, only cut parallel to the surface.
http://i65.photobucket.com/albums/h216/Ray_GTI-R/PC/Dscn2809a.jpg. Another lesson - the IHS alone weighs 18 grams. That's a little less than two £2 coins - which is quite heavy, really.
b) core and circuit board cleaned. All the original thermal compound removed and the circuit board cleaned of the black glue that fixed the IHS. Lesson - the thermal compound was now fried. It was still there but felt as crisp - as though it had dried-out.
c) & d) Lapped surface of the IHS. Confirmed - the IHS definiteely IS a copper slug plated with a silvery coat.
e) IHS positioned over the core without thermal compound. Lesson - there is quite a big gap between the edges of the IHS and the circuit board which MUST be filled with a partially compression-resistant material before fitting a heatsink to the CPU using the standard Intel socket 478 easy-clip heatsink/fan. This gap would be even bigger with thermal compound in place, although after compression-fitting of the IHS, the gap would be almost this size.
There was little hope that this CPU would EVER work again. It had lost it's HyperThreading ability. My bad cuts into the circuit board had actually removed a thin layer of the copper circuit(s) along one side and some pins had been very slightly bent during removal of the IHS. The glue holding the IHS in position proved to be very soft but exceptionally good at it's job. An unofficial article by ex-Intel employee who had engineered an IHS removal tool for Intel noted that to make the glue easier to remove required heating the entire unit. Since there was no mention of a target temperature I ignored it - probably not a wise decision given the damage I caused when I removed the glue).
Before attempting to fit the reworked CPU I took readings with the same equipment and a brand new equivalent processor to compare later - in case the reworked CPU managed to fire up that is. The baseline readings noted, it was time to see if the CPU was any good.
New thermal compound was applied to the core and the IHS glued back using ordinary silicone gel. The IHS was left in place over the core for 30 minutes under moderate pressure (3 rubber bands ... very high-tech!) then slotted it into a motherboard (it did fit!) and the standard heatsink was fitted. The plan was to leave it overnight for the silicon gel to set but impatience got the best of me.
Power on. No sparks, no smoke, POST completed! I checked the BIOS temperature ... same reading as the motherboard, so it wasn't overheating.
Progressed into Windows (98SE - the CPU would previously only work with this OS) and checked with CPU-Z and Sandra. The CPU showed all the right signs of good health. Started Windows XP Pro expecting the usuall BSODs from previous experience with this CPU. XP Pro fired up no problem.
And most amazingly of all ... HyperThreading was working once more.
Powered off and left it overnight to ensure that the silicon gel would set properly.
Restarted OK next day and took some more CPU-Z, Sandra and 3dMark results which were 100% OK. Felt brave enough to perform some burn-in (destruction) testing and still the CPU was fine. It was better than fine in fact ... the CPU temperature readings at 100% load were remarkably cool although the ambient temperature wasn't the same as the benchmark. At idle, the temperature was unbelievably cool (I use an Intel Extreme Edition cooler). I am waiting for the ambient temperature to rise back to the benchmark before jumping to any final conclusions.
Ambient equalised, CU still stable and 100%. Temperatures comparible with stock CPU so lapping the IHS didn't seem to be of much benefit. Ran Sandra burn-in for an hour , all OK.
Now I'll try overclocking - just trying to destroy this CPU now ... 2.6ghz OK. 2.9OK but unreliable. 2.95ghz failed to POST. 2.92ghz POSTED OK but didn't load XP Pro fully - Explorer hung. 2.91ghz loaded XP Pro, slowly, but failed 3dMark - CPU temperature OK and Sandra burn-in OK but weird fails (Found new hardware that hadn't been changed etc). These might be signs of RAM overload so I'll try my super-duper CL2 RAM.
Started again, with the fast RAM fitted, at 2.91ghz. Started XP Pro very slowly & another weird fail. So I've hit this reworked CPU's maximum reliable at just on 2.82ghz (235mhz FSB) at standard CPU voltage. I know that the fast RAM works at 266mhz FSB (1066mhz) from a previous Conroe motherboard test. My testbench motherboard does not permit voltage increases, so refitted standard RAM, reset to standard (2.4ghz) and re-tested OK. Retested at 2.82ghz OK. Fingertip checks of the heatsink, passive Northbridge heatsink, RAM, video RAM & GPU heatsink confirm they are all running cool. Retested over the next two days - rock solid!
The next time I get a motherboard to test and it has BIOS voltage adjustment I'll really push this CPU. Meanwhile - it's back to normal use - cool, low-power and potentially overclockable at standard voltages. With Hyperthreading available. Yeeeha!
The mere fact that the CPU has Hyperthreading again has made this entire exercise very worthwhile. The rest is truly a bonus.
FWIW - I strongly advise anyone who tries a CPU rebuild to do the research & understand it, get the right tools & materials FIRST, get anti-static protection (as ever) and go for it!