Moving forward from our air cooling results on to the more exciting aspect of this journey, we turn the cooling position over to a tired old Chilly1 single stage phase change cooler that requires a re-gas and re-tune very badly. The heat output of these new quad core processors when heavily overclocked is no match for this phase change but we will do our best with it anyway.
For those that don't know, a phase change cooler uses a compressor to compress a gas into a liquid that is very cold because it is compressed. This liquid is then fed into a small evaporator that makes direct contact with the processor and cooling it in the process. Phase change coolers can be bought retail from companies like Asetek with their VapoChill and Prometia with their Mach II/GTs. The Chilly1 phase change we are using is a privately built unit that would be very much comparable to a VapoChill phase change in its current state. We will start with some photos of the unit and the move on to the insulation that is required and finally some photos of the setup during testing.
For those familiar with phase change cooling, there is nothing too spectacular about our unit. It is very basic with just an on/off switch and no control software. It is not designed for 24/7 use and simply used for benching. It was initially tuned for 65nm dual core loads so desperately needs to be tuned for the higher output quad core processors of today. Even still, it should keep temperatures below -20C under load we figure on this Phenom II and that should be plenty to squeeze a good pinch more out of the processor from our air results. As mentioned though, before we get into any sub-zero temperatures, we need to insulate. Preventing condensation from forming is the key to any sub-zero computing.
To be truthfully honest, it took us about four and a half minutes to go from a bare board to a fully insulated setup. Aside from the single capacitor towards the top edge of the motherboard and the socket, no other cut outs are necessary. A simple rectangle armaflex layer with the PWM capacitors cut out is all this board really needs. We also put a small piece of armaflex tape over the cam end of the socket to ensure a complete seal against the CPU integrated heat spreader since the armaflex layer doesn't cover it. The entire motherboard back side is insulated in our test bench setup so absolutely no warm air will be near the underside of the motherboard causing condensation. Overall we are quite confident that this insulation job will be 100% adequate to do its job, despite the absolute minimum amount of time it took to setup. It isn't often these days that insulating is this easy but it certainly won't get any easier than the GA-MA790GP-DS4H. We will now finish off with a couple photos of the setup in action.
Our setup today is pretty straight forward and simple, aside from the phase cooler slapped on the CPU. As we can see though there is nothing really fancy with this setup, and only a couple fans providing the cooling for the stock motherboard. It seems about the time that we looked at some numbers and saw what this processor can do at sub-zero temperatures. We will start with the less demanding benchmarks and work our way down to tougher, multi-core, benches.
CPU-Z Validation
click for full size...
CPU-Z Validation | Hexus PiFast
click for full size...
 |
Super Pi 1M
click for full size...
 | Super Pi 32M
click for full size...
 |
WPrime - 32M
click for full size...
 | WPrime - 1024M
click for full size...
 |
Aquamark 3
click for full size...
 | 3DMark 01
click for full size...
 |
3DMark 03
click for full size...
 | 3DMark 05
click for full size...
 |
3DMark 06
click for full size...
 | 3DMark Vantage - Performance w/o PhysX
click for full size...
 |
Going from our air cooled results, these numbers are good but not jaw dropping. The strength of the phase change unit accepts the primary blame and shows this when going to multi-core benchmarks. The heat load produced at these voltages with these elevated frequencies is just too much for this phase setup. The WPrime results are the most telling with the difference between WPrime 32M and WPrime1024M being 437MHz. This is because temperatures raise so much after a minute or two of four cores under load. Any more vCORE and wPrime 1024M was into positive temperatures rather quickly. The 3DMark results from 01, 03, 05, and AM3 are again quite nice because they are single core benchmarks that allow the phase change to maintain temperatures of -30C under load. 3DMark 06 faired the best of the multi-core benchmarks but Vantage did quite poorly only being able to squeeze out a run at 4256MHz.
The vCORE we imposed for these phase results was 1.70v set in the BIOS. Some of the multi-core benchmarks needed that voltage to be lowered to run stable as heat was causing issues. We have mentioned a couple times that the phase setup we have was idling at -50C and up to that temperature, this system behaved beautifully. We were able to clock the HT and HTT buses without issue or complaint. There are a lot of reports that these chips won't be able to clock either once we go sub-zero but up to this point, our sample seems to be just fine. Let's now see what happens when temperatures are no longer an issue with LN2 on tap.
