As a first one, stock cooler was tested. Stock cooler was the only cooler, where the set up of fan speed rotation was not adjusted manually with regulator, but was made with ATI Tool software application. Resulting speed rotations were fixed gradually at 1500, 2600, 3500 and 5000 rpm.

Consecutive table and graphes recapitulate measured…

As you can see, at minimum rotational speed, stock cooler is not able to cool the graphic card in satisfying way, not even in idle mode. And actually most users, not only demanding ones, are able to accept only minimum rotational speed, not applicable for common using. Temperature of graphic chip of not-overclocked graphic card was after burning high. After overclocking the graphic card, test has to be stoped for registered critical temperatures. Silence and performance did not meet this time.
As quite impressive, I would consider recorded temperatures, when the fan was running full tilt with maximum rotational speed (5000 rpm.) it had no problems with cooling even after overclocking. But I would like to know somebody, who would like having this weird combination of helicopter and tractor by his side…or how to call this noisy thing. So again for common using, not usable. For those who do not want to buy a new cooler, compromise solution would be to regulate rotational speed to moderate values. Althought this compromising will not last a long time, as this kind of cooler is suitable only for not-overclocked graphic cards. I admit that also voltage regulators were not suffering, stock cooler was able to cool them in satisfying way, which would be other positive aspect.

After the first test, I quickly changed to Zalman VF900-Cu and repeated the whole procedure. This time, rotational speed was regulated manually in range recommended by manufacturer. There was only minimal deviation + 50 rpm, during real measuring with control of exact voltage with digital multimeter.

Zalman VF900-Cu is a different cup of coffee. This cooler is able to cool not-overclocked graphic card on a very good level. I do not count now, excessively high temperature recordered, when its fan was turned down, but that was more joke, that attempt to characterize Zalman cooler. VF900 is able to cool with minimal speed rotation, at nearly inaudible noise level when the test was run in the idle state of graphic card. In the idle state, there is only 1°C difference in temperature of graphic core, comparing speeds of fan rotation, when it is in minimal and maximal load. And there is only 3°C temperature difference between minimal and maximal fan speed rotation in full load, so it has no meaning to let the fan run full tilt.
Situation is the same with overclocked graphic card; only temperatures are 15 °C higher. There is notable difference in temperatures only when the cooler is set to low speed rotation, when VF900 is becoming to loose the breath.
Difference between Zalman and stock cooler comes, when temperatures of supply regulators are recorded. VF900 is small and its distance from MOSFET-passive is noticeable at resulting measured temperatures, which are slightly higher comparing to stock cooler. However, recorded temperatures of supply regulators were acceptable.

Third cooler in rank is GX815, Zerotherm VGA high-end cooler. Here the regulation of fan speed rotation was rather problematic. It is not possible to make changes in its own thermoregulation, during warranty period of cooler. Therefore, in table below, you can find for each measurement two values representing rotational speed: first value concerns speed in high load and second in idle load.

Differences in constructions of Zalman and Zerotherm tested coolers are very very small. Anyway in passive mode, there is a gap between these two - Zerotherm is winner here. But for the rest, VF900 shows better resuts.
There is notable increase in temperatures concerning descending rotational speed. It is effect of this particular fan, which is able to cool fins in better way at high rotational speed. On the other hand with rotational speed in low mode, its efficiency rapidly decreases. Looking at the graphics, you can see trend of decreasing temperatures under higher angle.
In the idle load, trend is, higher temperatures at lower rotational speed – regulated by thermistor. There are certain advantages, as in the idle load, low rotational speed is sufficient, and temperatures are not any high. But the thing is that noise is affected remarkably, and that is a drawback of GX 815. You’d better keep it in check – with lower rotational speed.