Testbed and Methods
The tests of all cooling solutions were performed in a closed system case only in identical testing conditions. The testbed was built with the following hardware:
- ASUS P5B Deluxe/WiFi-AP mainboard (Intel P965, LGA775, BIOS 1101);
- Thermaltake Extreme Spirit II chipset cooler (~2500rpm);
- Intel Core 2 Quad Q6600 processor 2400MHz overclocked to 3358MHz (4x266MHz FSB, 2 x 4096KB L2 cache, 1.4875V, Kentsfield B3);
- Scythe Infinity CPU cooler (120mm fan at 1100rpm);
- Fanner 420 (Stars) thermal interface;
- 2x1024MB DDR2 Corsair Dominator TWIN2X2048-9136C5D memory (SPD: 1142MHz, 5-5-5-18, 2.1V);
- Hitachi HDT725032VLA360 HDD (SATA-II, 320GB storage capacity, 7200rpm, 16MB cache, NCQ)
- Optical drive: NEC ND-4571A DVD-burner
- ASUS ASCOT 6AR2-B Black&Silver system case with two 120mm Sharkoon Luminous Blue LED (~980rpm) system fans and a 120mm Cooler Master fan (~800rpm) a on a side panel
- Enermax Galaxy EGA1000EWL 1000W power supply with a 135mm intake fan at ~850rpm and 80mm outtake fan at ~1650rpm.
The tests were performed in Windows XP Professional Edition SP2 operating system. We used DirectX 9.0c (released July 2007), ForceWare 93.71 and Catalyst 7.6 graphics card drivers. The graphics cards were warmed up by running the Firefly Forest test from the synthetic 3DMark 2006 benchmarking suite 10 times with disabled full-screen anti-aliasing but with enabled 16x anisotropic filtering.
The graphics cards temperatures were monitored with RivaTuner version 2.02 utility. Each cooler was tested at least two times with a stabilization period of 13-15 minutes between test cycles. If the results obtained in a repeated test were more than 2o C different from the previous ones, the tests were performed again until we got fine precision results. The room temperature during this tests session was stable at 26o C (marked with vertical red line on the diagrams).
The noise level of the tested cooling systems was measured at 3cm, 1m and 3m distance according to our methodology described in the previous articles. The background noise of the system case measured at 1m distance during our test session equaled 34.1dBA.
The average fan rotation speed of all testing participants in different modes is listed according to the monitoring data, and not according to their default specs. Reference graphics card cooling solutions are marked as “Stock cooler” on the diagrams below. We used GeForce 7900 GS 256MB with the standard cooler and overclocked to 576/1720MHz chip and memory frequency respectively. It has already been shown on the photo above. Another graphics card we used was Palit Radeon X1950 GT 512MB with a non-standard default cooling system onboard:
This graphics card’s cooler is designed with an aluminum blade heatsink cooled down by a 92-mm fan covered with a plastic casing:
In Auto mode the fan runs at 40~45% of its maximum speed. The maximum frequencies when Palit Radeon X1950 GT 512MB remained stable were 587/1382MHz for the GPU and memory respectively (500/1200MHz nominal speeds).
Besides the default graphics card cooling solutions we have also included the results of a pretty popular Zalman VF900-Cu LED cooler on copper heatpipes (in two operational modes: at 1800rpm and at maximum 2650rpm). We also included the results for Arctic Cooling Accelero S1 (in passive mode and with one 120-mm fan from Scythe running at its nominal speed of 1200rpm). Unfortunately, we couldn’t test Arctic Cooling Accelero S1 on Palit Radeon X1950 GT, because the cooler legs were hitting against the PCB textolite way before the base touched the GPU heat-spreader, so there was no contact at all. It is a similar situation as the one we have just described above with the Radeon HD 2600 XT and Gigabyte V-Power cooler, although in that case it was the GPU frame that prevented the cooler from fitting properly.