Problem with using overclocks is that this can be notoriously unreliable.
Some CPU's will go right to a limit, and then won't budge until you take another 5C off their temps, while others will scale quite gracefully, allowing an extra 10MHz per 1C drop or so.
The way BillA's has tested it is pretty much what we want. About the only thing missing from BillA's test data is a firm statement of the wattage output of various CPU's when over-clocked and over-volted. With that in hand, it's then fairly easy to say block X will cool better by Y degrees.
System thermal probes are difficult to use properly. By that I mean they are never accurate with respect to absolute temperatures, and the linearity of their temperature scaling is not guaranteed either. At best, they give a decent indication of if something is being cooled better by one block over another, but little more.
Better thermal transfer efficiency (characterised by a lower C/W) ALWAYS means a lower CPU temperature, which in turn correlates into the potential of a better overclock. Exactly how much that is is dependent upon a per-CPU basis.
I have people who installed my block over a Maze 3 reporting anything from as good as 100MHz gains on P4's, to as low as just 20MHz gains on AthlonXP's. Of course, who's to say the reasons for each result? Was it really cooling related. If a better overclock resulted, then clearly it is, but the size of the extra overclock may be limited by other factors, such as voltage and FSB stability.
Overclocking potential is just too messy. Further, unless the room temperature and coolant are held at exactly the same temperature all the time, then this may mean that CPU and or motherboard are experiencing inconsistent temperatures which may all affect the end-overclock.
What I think is better here is an article that details the peculiarities of overclocking, the variables involved, how a CPU's overclocking potential is linked to CPU temperature, how motherboards are not accurate within themselves, let alone with respect to each other, and then using BillA's data for some different pumps, radiators, blocks and fans (on the rads) results in a coolant temperature of X and a die temperature of Y, and give some good predictions on what setups will work well.
|