Test Results: New-Generation Coolers
In order to analyze thermal efficiency of the GPU coolers under review, we have followed practically the same method principles that are used in our CPU cooler tests. That is, the primary data are again temperature readings of the built-in thermal diode, what's different is the thermal source (now it's VPU NVIDIA G92), the base platform (NVIDIA GeForce 8800 GT graphics card), and a set of test applications.
We've found out an interesting fact in the process of preliminary analysis of thermal characteristics demonstrated by the GeForce 8800 GT in 3DMark06: complex, graphically intense scenes "Canyon Flight" and "Deep Freeze" (SM 3.0) show GPU heat release of about 55 W, while relatively simple tests "Fill Rate (Multi-texturing)" and "Pixel Shader" managed to raise VPU to 70 W. So, instead of "Return to Proxycon", "Canyon Flight" and "Deep Freeze", which were used in our GPU shootout last year on the ATI platform, our today's contenders were heated with a looped combination of Fill Rate (Multi-texturing) and Pixel Shader tests.
The graphics card operated at 740/1728/2000 MHz (ROPs/Shaders/RAM): when the stock cooler was set to 2500 RPM (that is when the noise was still on an acceptable ergonomic level, while higher speed was too noisy), it was the maximum acceptable mode for our sample of the GeForce 8800 GT. Higher clock rates resulted in instability. So we've chosen this mode for our shootout of GPU coolers.
We have also inspected thermal operating modes of video memory and VRM: we have measured the temperature on M1, M2, and M7 memory chips as well as on the following transistors—Q17, Q18 (in VPU voltage regulator circuits) and Q24 (in the video memory voltage regulator circuit) with calibrated electronic thermometers, ensuring engineering accuracy of measurement with the consequent statistic analysis of the test data.
Here is our testbed configuration:
- Motherboard: ASUS P5AD2-E Premium rev. 1.05
- CPU: Intel Pentium 4 550
- Graphics card: NVIDIA GeForce 8800 GT
- OS: Microsoft Windows XP
In order to emulate a VPU thermal load close to maximum, we used looped Fill Rate (Multi-texturing) and Pixel Shader tests from 3DMark06. Each cooler was tested in three 60-minute stages, the environment temperature (innards of the half-open testbed) was kept at 30B°C. And finally, we used the RivaTuner utility to monitor VPU temperature.
Each cooler was tested with the stock thermal interface.
Background noise level: 18 dBA.
Well, the new GPU coolers demonstrate an evident advantage over the stock cooler of the first revision in all parameters—GPU cooling efficiency, and accompanying ventilation of VRM elements, and especially memory cooling. An evident leader today is the Thermalright HR-03 Rev.A with the Scythe Minebea fan (although the reverse orientation of the HR-03 Rev.A is less efficient—the defeat at 800 rpm is already significant!) The Zerotherm Hurricane HC92 Cu 8800 also demonstrates excellent results. However, with one negative reservation—the fan is too noisy at maximum speed, so the ergonomic mode is limited to 2600 rpm.
These GPU coolers expectedly form the best tandem with the GeForce 8800 GT—we managed to raise operating frequencies of the graphics card to 817/1890/2150 MHz with these coolers! To all appearances, that's a physical Rubicon, which can be crossed only using a voltmod. Two other coolers are less efficient. Nevertheless, their results are also respectable—800/1782/2150 MHz! :)
Write a comment below. No registration needed!