Performance
Testing SetupRemember, it's a closed loop system and the GPU block is the first block in the system after the heat is removed at the heat exchanger so its heat will be passed on to the CPU block. So I left the Ti4400 at stock clocks (275/550) while testing the CPU performance of the MCX462-U block.
Here's the system specification.
• Swiftech QuietPower FS020 Watercooling System in H20-C-W40 configuration
• Swiftech MCW462-U CPU Waterblock
• Swiftech MCW40 GPU Waterblock
• Swiftech MCX478 air cooler
• Lian Li PC78 with 4 x 80mm front intakes and 1 x 120mm rear exhaust
• Intel Pentium 4 2.26 'Northwood' @ 17 x 133MHz and 17 x 166MHz (2.26GHz and 2.833GHz)
• ABIT IT7-MAX, Socket 478 Intel Pentium 4 DDR Motherboard
• 256MB Samsung PC2700 DDR SDRAM
• MSI GeForce4 Ti4400 128MB
To test idle performance, the systems were left idle in Windows XP after a fresh reboot and default tasks running for 30 minutes. Ambient temperature was ensured to be 22c during testing both systems via a highly technical heat extraction technique called "opening and closing the window and using a pocket thermometer to test air temperature". Not very scientific I know but the pocket thermometer verified the room air temperature to be 22c over the 2 days of testing.
The 4 80mm intake fans and the single 120mm rear exhaust fan in the Lian Li case (the 120mm exhaust is very close to the CPU area and responsible for a lot of my air cooling performance since it removes heat quickly) were run at 12 volts throughout testing. I run them all on a fan speed controller and case temperature was measured to be just a shade under 28c on the day of testing in the Lian Li and just over 26c in the Swiftech case on that day of testing.
CPU Waterblock performance
Performance was tested using SETI over a 2 unit period (just under 6 hours of load per test) and then a full 4.5hr DivX encode of a pair of films, one per cooler. 10 hours of load per cooler to arrive at the numbers. Idle performance is explained above. Let's see how they do.
At stock CPU clocks, both cooling solutions do an identical job at idle. There's enough cooling capacity in the air cooler to do an identical job. Under load however the watercooling shows off its muscle. Rather go into detail as to why the watercooling performs better here, lets look at the overclocked CPU case and use the same reasoning for those results.
The same effect here, just exaggerated compared to the stock CPU case. It's a fact that water cooling is able to remove more heat, more efficiently than air cooling. Water can absorb heat from its surroundings better than air so even with a powerful fan, using water to draw heat away from a heatsink (which is effectively what the waterblock is) will be a better cooler. Provided you can convert the heat well at the heat exchanger, a watercooling system will almost always outperform an air cooled system.
Sub 50c load doing DivX encoding, one of the most stressful CPU bound tests on a modern PC, at 1.65V is highly impressive, over 10c less than the air cooler. The manner in which it goes about this cooling is what impresses even futher. It was almost silent and the noise from my noisy SCSI drives was more than enough to mask out the slight fan noise. The pump is silent.
Impossible on my air cooling previously, the Swiftech was also able to give me 99% stability at the following speed.
Load temperatures doing SETI at this speed (the main task on my CPU these days) were never more than 55 celcius. The CPU would be over 10c higher on air cooling and stability was very hit and miss.
So we know how the watercooling achieves its impressive performance on the CPU, can it do the same on a GPU in the same cooling loop?
GPU Performance
Since I have no way of directly monitoring GPU temperature, GPU performance will be illustrated by a series of comparisons. A variety of games were tested, each puts different stress on the GPU and therefore the maximum stable clock will be different when playing each game. The new maximum stable limits were tested and we'll see what kind of clock increases we can get. The card is a relatively decent clocker to start with and the stock cooler was attached strongly and had a good layer of thermal material out of the box direct from MSI.
Where benchmark tests are involved (3DMark 2001, Aquamark) the core clock reflects the ability to loop the benchmark for a long period of time, not just a one off run of the program in question. Core clocks for the game tests (UT2003 and Quake3) were obtained by actually playing the game for an extended period of time (over an hour).
3DMark was most receptive to the cooling solution and I saw a 19MHz increase in core clock to 334MHz (275MHz stock clock remember). UT2003 was least receptive and seems to choke an NV25 card quite badly with its graphical requirements in terms of GPU power. However the increase was there over air cooling which is what we wanted to see and of course it was a lot quieter than the noisy fan on the original MSI cooler.
So you get the same style of benefits on a GPU block as with a CPU block and for the same reasons.