Testbed and Methods
We tested the cooling system Gigabyte 3D Galaxy using the following computer:
- Mainboard: ABIT AG8-V (i915P), LGA775, BIOS v.2.1
- Processor: Intel Pentium 4 540 3200MHz, 1024KB, 800MHz FSB, LGA775 (Prescott, E0)
- Graphics card: PCI Express GeForce 6600 256MB Sparkle (375/500MHz)
- Memory: 2x256MB DDR PC3200 400MHz Patriot XBLK (2.5-3-3-5)
- Hard drive: 200GB SATA Seagate Barracuda 7200.8 (3200826AS) 7200rpm, 8MB buffer
- DVD±R/RW & CD-RW NEC ND-3520 A firmware v.1.05
- In-Win S508 system case + Thermaltake-W0009 420W power supply + two 80mm system fans from Zalman (~1750rpm, 7V)
- Monitor: LCD DELL 1800/1FP UltraSharp (1280x1024, DVI, 60Hz)
The OS and drivers: Windows XP Home SP2, DirectX 9.0c, NVIDIA ForceWare 76.50.
There’s no thrill in testing a liquid-cooling system on a non-overclocked processor, even though it is the hot Prescott core. So I sped up the Intel Pentium 4 3.2E to 4100MHz to increase its heat dissipation (to do the overclocking I raised the core voltage from the default 1.3625V to 1.4625V):
Running a little ahead, I want to say that the Gigabyte 3D Galaxy made it possible for this processor to work at 4170MHz with 1.4825V voltage, while neither a GlacialTech Igloo 5100 PWM nor a super-effective Scythe Ninja could ensure a stable operation of the same CPU at frequencies above 4100MHz.
I followed our traditional testing methods: the processor was warmed up with the latest version (1.7.1) of the S&M utility:
The thermal throttling feature of the Intel Pentium 4 3200MHz processor on the Prescott E0 core was controlled with the ThrottleWatch version 2.02 utility.
I tested the 3D Galaxy at three speeds of the 120mm fan and performed three warm-up/cool-down cycles in the “long” mode of S&M (60 minutes long) at 100% CPU load. Considering that the S&M utility creates a very high load on the CPU, untypical for most applications, I also used the Super PI utility that calculates the pi value up to 32 million decimal digits which takes about 24 minutes on a fast enough processor (I launched two copies of the program to load the both virtual processors). The end result was the average of the three test cycles. It was rather hot here during my tests, so the room temperature was about 27°C. All tests were performed with the system case closed and standing in an open niche of a computer desk. The CPU temperature and the speeds of the fans of the cooler and of the cooling system were controlled through the ABIT µGuru utility during the tests.