So, I’ve had my new 2600K system up and running good for a little over a week now and decided to do some exploration on water/core deltas. One of the problems I’ve had with past CPU block testing on the Q6600 was the fact that it would not hold if ambients were allowed to fluctuate. This in turn required that ambient temperatures were fixed (as much as possible) and that testing was done based on water/air deltas. Unfortunately that sort of testing introduces a fair amount of error with the radiator/fan combination. Minor voltage fluctuations or other oddities can make fan rpms vary which varies the water/air delta. In addition any sort of radiator dust buildup between tests can also affect results to some degree. For me it was more of a struggle to maintain any sort of smooth ambient temperature. The best I was capable of doing was using an air conditioner in a close room that varied ambients up and down over a 2C range. Over enough logging of data that would smooth out the bumps, but I still was never very satisfied with the controls.
Fortunately, things seem to have changed for the better when it comes to the new i7-2600K series. I ran a few tests purposely varying the ambient levels. I’m running about a 2C water/air delta with the quad radiator, but my house temperature was purposely fluctuated such that I had about an 8 degree change in ambient and water temperature. The resulting informal tests produced nearly identical results which is great news for consistency.
While there are some important items to acknowledge in water/core type testing, it should provide a substantial improvement to testing consistency.
Here are those quick couple of tests:
That’s good news and should mean a rather simplified means to isolating block or TIM performance differences. Of coarse pressure drop is important to account for radiator thermal performance differences, but that’s easy enough to do and eliminates several variables.
I’ll see what I can do to round up a few CPU blocks to see how they fair on sandy bridge..
Still exploring and setting up for some CPU block testing. Here are some of the Dallas one wire probes I’m deploying via the Crystalfontz.
Exploring loading programs, here is OCCT vs Prime 95 custom small 8K FTT:
I find the logged charts fairly interesting. They are a bit of a pain to create between the two programs because there are minor time errors on both Realtemp and the Crystalfontz. Because of these minor errors, it makes matching them up a rather tedious Vlookup function combining effort. I wish I could just push one button and get the above, but I can’t…it takes work. I’ll have to work on some excel templates to make it less painful.