Results & Temperatures
The results are generally quite encouraging, possibly even a little better than I expected. Connecting the components in series doesn’t seem to have as much impact as I feared. The 700lph is more than enough to keep the water moving fast enough to prevent it warming up too much as it passes through the various coolers.
: All temperatures are above ambient and the machine is placed under a desk in a corner with average ventilation at best. Optimal positioning and access to cool air (eg an open window) lowerd the temperatures but didn't change the relationship between air-cooled setups and the water-cooled setups significantly. The case was closed for all tests.
The CPU temperatures are quite acceptable for a watercooled system.
CPU-temperature (above ambient) vs. Clockspeed (MHz)
Despite the components being cooled in series, the temperatures are quite ok. For example KT400 chipset owners should benefit more as KT400-based motherboards tend to run hotter than this nForce2 board.
Chipset-temperature (above ambient) vs. Clockspeed (MHz)
I initially plotted this against MHz FSB, but this doesn’t make any more sense than plotting it again CPU MHz or any other criteria. I can only attribute this to the fact that the "system" temperature is some kind of averaged temperature over several components. When I aimed a 12" fan over the system the reported temperature dropped immediately whereas the temperature of the CPU stayed the same. The temperature probe I inserted under the chipset-cooler was also showing me temperatures of around 10C to 12C above ambient no matter what configuration I was currently testing.
The AGP Card
The FX5200 card doesn’t seem to produce too much heat, but I could find no way of reading the temperature in software, so I had to fix a probe to the GPU cooler. In any case, the Plug&Chill made all the difference on this AGP card.
VGA-temperature (above ambient) vs. Clockspeed (MHz)
Performance benefits - Benchmarks
In terms of overclocking the system and the resulting benchmark scores the Plug&Chill does quite well. It helped achieve better results, for the most part by keeping the temperatures lower and less variable because it can remove significantly more heat than air.
After running SiSoft Sandra a few times I took averages of all the results, ignoring the highest and lowest of each of the 5 test runs:
XP2500+ scores (default speed):
- CPU: 6840/2775
- MMX: 10148/10840
- Memory: 2403/2238
~XP3500+ (2400MHz) scores:
- CPU: 8988/3645 (+31.3%)
- MMX: 13303/14239 (+31.2%)
- Memory: 2849/2701 (+18.5%)
The same testing method was applied to 3DMark 2001 SE 330 and 3DMark03 Pro:
FX5200/64MB @250/333 on XP2500+@1833MHz(default):
- 3DMark2001SE 330: 4436
- 3DMark03 Pro: 1344
FX5200/64MB @287/450 on ~XP3500+ @2400MHz:
- 3Dmark2001SE 330: 5772 (+30%)
- 3Dmark03 Pro: 1822 (+36%)
I would take these 3DMark scores with a pinch of salt as I had tremendous problems getting this card to even complete the benchmark at any setting. Also take into consideration that the scores are heavily driver dependent. Both tests were performed with the same drivers (nVidia 43.25 and DX9.0b) so the relative increase in performance should still paint a fairly accurate picture of the overall improvement.