An experiment to determine the efficacy of CPU back cooling. – Joe
SUMMARY: Opening up the CPU socket yields additional cooling but not as much as I thought.
You’ve all seen sockets like this – lots of little components arrayed around the bottom with a blue eye looking up at you. However, this one is a little different – if you look very closely, you’ll notice that all the active components are close to the edge of the socket. There are solder pads and lead holes that are unused.
Outlined in RED above is the “safe” area where it might be possible to cut the motherboard without impacting circuitry. So, I used a dremel bit in my drill press and proceeded to do some cutting:
I did this in small steps, continually checking to see if it was working. First, a 1/16″ hole in one of the unoccupied lead holes. I ran the board, it posted, then back to the drill press, more holes, then a routing bit, etc. Cutting out the in-socket diode caused no problems – it now reads 0 C. After about 8 trips back and forth, I stopped here for fear of screwing the pooch.
My objective here was to test the idea that air cooling the back of the CPU will drop temps. Question is by how much? The old timers (3 years ago) remember Celeron sandwiches, so they understand the concept.
To test it out, I ran Prime 95 without any cooling and with a small squirrel cage fan aimed directly at the socket hole. After numerous runs, the best I can do is drop CPU temps, measured at the CPU case top with a Glaciator II thermocouple in the base, by 0.5 C.
This is about the same cooling benefit we found by aiming fans at the back of the motherboard, as found HERE.
Looks like the real cooling benefit from secondary cooling comes from cooling things like voltage regulators and capacitors around the socket. This whole thing surprised me – I expected more from it. I’ll continue to experiment and report back if I get better results.
Where I come down on this is that if it’s possible to open the socket, there is some cooling benefit and it’s “free”, so if motherboard manufacturers can accommodate this, why not?
I also think this nonsense about “airflow” in the socket from some heatsinks which direct air down onto the motherboard is just that, and attempts to “correct” results by insulating inside the socket does nothing. In socket diodes are not, and never will be, test instruments.
The motherboard I did this on is an engineering sample of Iwill’s XP333 – I have no idea if this is possible on the release version and absolutely do not recommend anyone attempting this.