Delidding Haswell Core i5s
Someone on Reddit or Discord or something, I’m not sure which, mentioned the thermal paste in Dell micro-form-factor machines not being so shit hot, and I thought “well maybe that’s why mine seem like they’re a bit toasty piled into the rack in summer?”.
Since I’m a fiddle-fingers with a Die Mate 2, a bunch of extra liquid metal left over, and a whole stack of USFF machines not doing anything, I figured why not.
Cracking one of the spare machines open, I found the thermal paste was properly dried out, cracking, and crumbling. I thought for sure we’d see some big gains there alone, but since I’m actually running low on Noctua thermal paste, rather than do some control tests to see how much just changing out the TIM would help, I decided to press on.
Pop the CPU out, put it in the Die Mate, push the lid off like it’s nothing. Used my fingernail and the edge of a credit card to get rid of the silicone RTV that held the spreader on, put a bit of nail polish over the capacitors like normal, spread some conductonaut over the spreader and the die, put it all back together. Put the Noctua stuff on top of the cleaned up heatspreader, put the heatsink back on and:
Package id 0: +40.0°C (high = +69.0°C, crit = +75.0°C)
Core 0: +40.0°C (high = +69.0°C, crit = +75.0°C)
Core 1: +40.0°C (high = +69.0°C, crit = +75.0°C)
Core 2: +38.0°C (high = +69.0°C, crit = +75.0°C)
Core 3: +38.0°C (high = +69.0°C, crit = +75.0°C)
Package id 0: +39.0°C (high = +69.0°C, crit = +75.0°C)
Core 0: +37.0°C (high = +69.0°C, crit = +75.0°C)
Core 1: +36.0°C (high = +69.0°C, crit = +75.0°C)
Core 2: +38.0°C (high = +69.0°C, crit = +75.0°C)
Core 3: +38.0°C (high = +69.0°C, crit = +75.0°C)
Those are two identical machines (roughly), running roughly identical loads, on the same shelf in the same rack, after some 30 minutes to equalize. The fact that I would have to tell you which one is which does not bode well for getting the huge gains I was sure I’d get from the TIM alone. The fact that if you guessed which was which, you’d probably be wrong is hilarious - the top one’s the delidded+relidded unit.
Considering the massive gains in idle and load temperatures I got on my Kaby Lake I’m a little surprised at what little difference it made. I guess it speaks to the fact that the Kaby Lake i7 is a much hotter chip, and the TIM Intel used under the heatspreader was actually up to the task once upon a time… it was just not optimal for the 7700k.