SUMMARY: Thermosyphons – one tube at least as good as two.
In our quest to reduce costs as much as possible, we decided to remove one of the stainless steel tubes leading to the condenser – there is no reason for two tubes as one adequately sized tube will do just as well. I should note that this tube is flexible so that it can be oriented to the CPU to be cooled:
We are in the process of sourcing parts – I thought you’d like to see a condenser we are testing to our spec:
We are using an all-copper condenser – this might make a good watercooling rad!
The Thermosyphon shown above (spec’d for 200+ watts) was tested on an Acorp 4S845A motherboard with a modified P4 1500 to read CPU case temps. For testing, I used 2 NMB 92 mm fans (#3610K4-04W-B50 vented) spinning at 3200 rpm.
TEST RESULTS – Acorp 4S845A Motherboard
CPU Case Temp
3210 rpm, 61 dBA²
¹MBM on-die temperatures.
This places results at the top of P4 air cooled solutions I’ve tested to date (HERE), with the added benefit of a cooling solution that will handle 200+ watts. One of the issues facing heatpipes is that their effective temp range is more limited compared to thermosyphons; as watts increase, the heatpipe solution is to add more and larger heatpipes – hence the size increases I saw at Computex (HERE).
As a further test, I mounted the Thermosyphon on an Asus K8V Deluxe with an AMD Athlon 64 3200+. At spec, it radiates 89 watts. I’m also showing results at idle and under stress at spec and overclocked 10% with Prime 95 using the on-die diode – I can’t attest that the temps are accurate, but the difference between idle, stress and overclocked are interesting:
AMD 64 @ 2.00 GHz – Idle
AMD 64 @ 2.00 GHz – Prime 95 (89 watts)
Now overclocked 10%:
AMD 64 @ 2.20 GHz – Idle
AMD 64 @ 2.20 GHz – Prime 95 (98 watts)
Not much difference between spec and overclocked! While I can’t say that CPU temps are accurate, it does calculate to a stress C/W of 0.18 at spec and 0.17 overclocked; these results are almost the same as for the two tube version (HERE).
The last test I ran was using a die simulator with a die measuring one inch square at 99.3 watts. This test resulted in a C/W of 0.12. I don’t have comparable results to the small die test as the heaters on my die tester finally gave up the ghost and it’s out of service for repair.
The further along we get, the more I am convinced that thermosyphon cooling is the next logical step to tame increasingly hot, high wattage CPUs at low noise levels and reasonable cost. In addition, the weight on the motherboard and CPU is greatly reduced compared to large heatpipe solutions, as the “business end” (condenser) is not tied to the socket or motherboard (drop tests give me the shivers).