Power Mac G4 cooling project: Phase 3 - Orientation swap


Hello and welcome to the third installment of the Power Mac G4 cooling project. If you missed the last two installments here are the links...

Power Mac G4 cooling project: Phase 2 - The baseline
Power Mac G4 cooling project: Phase 1 - The plan

In this installment I will be changing the orientation of my 120mm side panel fan and my 60mm CPU fan. Originally the CPU fan was oriented to draw external air into the case and onto the CPU heatsink. The Apple OEM heatsink for my Quicksilver did not have a fan on it; however, my third party CPU has two 40mm fans on top of it. With that in mind, I will orient the 60mm fan to expel the internal case air out of the case. The Quicksilver's CPU, RAM and GPU all live very close together. I suspect that expelling all of that hot air may bring the temperatures down. The results will answer that for us. The 120mm side panel fan is originally oriented to draw air out of the case. I will be changing it's orientation to draw external air into the case. I suspect that using the 120mm fan to draw air into the case may help the GPU temperatures; again, the results will answer that for us.

At the end of phase 2, I came to the conclusion that, on my Quicksilver, nap mode was enabled by default under OS X. With that being said I will not be performing an idle test without nap mode enabled.

I will note the difference in temperature taken, from the baseline temperature taken in phase two.

With all that said, onto the tests!

Ten minutes of idle. OS X 10.4.11:
CPU: 39°C, -3°C
GPU: 43°C, +1°C
RAM: 36°C, -3°C
Exhaust air: 33°C
Intake air: 36°C, +2°C

Ten minutes of idle. Debian Wheezy:
CPU: 38°C, -4°C
GPU: 43°C, +1°C
RAM: 36°C, -6°C
Exhaust air: 33°C, +1°C
Intake air: 38°C, -4°C

This very interesting already. Swapping the orientation did benefit the CPU and the RAM temperatures. There is a catch though; the benefit comes at the expense of the GPU temperature. A one degree increase is not large but this is just at idle. Once I actually start taxing the GPU with something like Quake III; I suspect that the increase in GPU temperature will grow larger.

I will start with the web browsing test. I am curious to see how the GPU temperature increase trends. I will have three tabs open, PowerPC Liberation, FrozenCPU.com and Newegg. I will bounce between FrozenCPU.com and Newegg just as I did in phase two.

Surfing under TenFourFox 31. OS X 10.4.11:
CPU: 48°C, -3°C
GPU: 46°C, +3°C
RAM: 39°C, -4°C
Exhaust air: 36°C, +2°C
Intake air: 38°C

Surfing under Ice Weasel 24.7.0. Debian Wheezy:
CPU: 51°C
GPU: 46°C, +3°C
RAM: 39°C, -4°C
Exhaust air: 36°C, +2°C
i Intake air: 39°C, +1°C

This keeps getting more and more interesting! The GPU temperature did slightly increase, going from a one degree increase to a three degree increase on both platforms. The RAM temperature also maintained it's decrease under both platforms. This is where it gets interesting, I managed to completely loose my CPU temperature decrease under Debian. This makes me start to question the benefit of this orientation swap. It looks as if, when pushed hard enough and long enough, the CPU temperature rises to the baseline temperature. So far I have gained ground on the RAM temperatures and CPU idle temperatures; while loosing ground on the GPU temperature and possibly the CPU temperature under higher usage.

Now lets see if I can erase the drop in the CPU temperature again by pegging the CPU at 100%. To do this I will perform the DVD rip test again. This is the same DVD rip test from phase 2 using using Handbrake 0.9.1. under Mac OS X.

The DVD rip results are...
CPU: 57°C, -4°C
GPU: 43°C, -3°C
RAM: 39°C -9°C
Exhaust air: 36°C
Intake air: 38°C, -4°C

Just when I thought this could not get any more interesting, it did. For the DVD rip we have a drop in the CPU and GPU temperature. We have a sizable nine degree drop in the RAM temperature. My theory on this is as follows. The CPU, RAM and GPU are very close to each other on the Quicksilver motherboard. The DVD rip is very taxing on the CPU. Out of all the tests I run this test is the hardest on the CPU. The CPU being taxed like that will obviously generate more heat. I suspect that the generated heat stays in that area. The RAM is next door to the CPU and the back of the GPU is right next door to the RAM. Now that I am immediately expelling the hot air with the 60mm fan I do not believe that the CPU heat is staying in that area as much as it did prior to the orientation swap. With that being said, the DVD rip is not very taxing on the GPU at all. The AVI playback tests and the Quake III tests will be more GPU intensive. Let us see what happens when I get the GPU to put out some heat.

The third test will be the playback of the resulting AVI file. I will use VLC under OS X and Debian. In addition to VLC, I will test mplayer in Debian.

OS X 10.4.11 VLC 0.9.10 full screen playback:
CPU: 43°C
GPU: 43°C, +4°C
RAM: 38°C, -1°C
Exhaust air: 33°C
Intake air: 38°C, +4°C

OS X 10.4.11 VLC 0.9.10 windowed playback:
CPU: 43°C
GPU: 43°C, +1°C
RAM: 38°C, +3°C
Exhaust air: 34°C
Intake air: 38°C, +5°C

Debian Wheezy VLC 2.0.3 full screen playback:
CPU: 43°C, -3°C
GPU: 46°C, +3°C
RAM: 38°C, +4°C
Exhaust air: 34°C, -2°C
Intake air: 39°C, +5°C

Debian Wheezy VLC 2.0.3 windowed playback:
CPU: 43°C, -3°C
GPU: 43°C
RAM: 38°C, -4°C
Exhaust air: 34°C
Intake air: 38°C

Debian Wheezy mplayer svn34540 full screen playback:
CPU: 43°C, -3°C
GPU: 43°C
RAM: 38°C, -4°C
Exhaust air: 34°C
Intake air: 39°C, +3°C

Debian Wheezy mplayer svn34540 windowed playback:
CPU: 43°C, -3°C
GPU: 42°C, -1°C
RAM: 37°C, - 5°C
Exhaust air: 34°C
Intake air: 38°C

Quake III Arena OS X 10.4.11:
CPU: 54°C, -3°C
GPU: 44°C, +1°C
RAM: 40°C, -3°C
Exhaust air: 36°C, +2°C
Intake air: 37°C, -1°C

Now it is time for my final thoughts on this experiment. I was honestly convinced that using the 120mm fan to draw air in; combined with using the 60mm fan to expel the CPU and RAM hot air would bring the temperatures down. I did not want to make such a claim without having data to back it up. That is where these tests come in.

I was wrong in regards to the effect of switching the 120mm fan's orientation. The tests, for the most part, had raises in the GPU temperature. Even though I am blowing cooler external air onto the GPU using a 120mm fan, the GPU temperature went up. I believe the reason for this is the fact, that by switching the orientation, I have given up a large form of exhaust. The heat generated by the GPU and the PCI area does not have the immediate exit it had before.

Switching the orientation of the 60mm fan however; did cause drops in the CPU and RAM temperatures. The tests, for the most part, confirm my theory about this area of computer. Immediately expelling the hot CPU and RAM air usually lowered the CPU and RAM temperatures.

The results are mixed. Since this has negatively affected portions of the system; I do not believe I can recommend this orientation swap as an improvement to a Quicksilver's cooling system. The cooling system in it's stock form does not perform poorly; therefore I do not see the sense in raising some temperatures for slight decreases in other temperatures. The mission of this project is to make it as good as I possibly can. I believe that this orientation swap misses that mark.

That brings us to a fork in the road. That is where to go from here. I could add a fan to the top of the airport bracket like Zen has in his Power Macs and test it's impact. I could also opt to replace all of my fans with Noctua or Antec fans, then test the impact.

I look forward to feedback. I hope you all have enjoyed this project so far.

2 comments:

  1. Nice experiment, really interesting. I'd like to replace the fans on my QS because they are a bit too loud, but it's not a massive problem.
    Nice job!

    ReplyDelete
  2. Thanks a lot for the research and the data that you have logged. IMO the best is to change the original fans for a Noctua or Scythe, and add another on the airport bracket like Zen did. would be great to test that under your protocols. Thanks.

    ReplyDelete