There are some issues with testing with XP diodes, no doubt. For the kind of testing that BillA and Joe are talking about, they are not suitable. In fact, there is no way to test performance of waterblocks with a processor reliably. You are going to have to move to synthetic setups to get any hope of reproducibility.
Since the XP diode readers seem to keep surfacing in this thread, I will now talk frankly about the diode readers that can be made from maxim ICs. There a few obstacles that can stand in the way of their use in testing, some surmountable and some not.
For one thing, can they be calibrated? I think that the best way might be something like the following: Take your CPU and connect it to the reader, then move that CPU into water of different known temperatures. You should be able to construct a reasonable calibration that way. Now comes the kicker: you'll have to code some software that is capable of accepting this calibration equation rather than just an offset if it isn't simply a constant. Do the calibration after removing and replacing the wires: does it change?
Next, carefully look at your lead wires from CPU to reader. You will find that their length and their gauge likely will affect the readings you get. You won't be able to notice this until you try testing a heatsink at a wide range of heat loads. You'll notice that if your leads are much too long then the relationship between deltaT and heat load will not be constant. I know that C/W values in temselves aren't particularly reliable, but you'll find if you shorten the leads then they become constant. The shift in the deltaT for having leads too long is easily as large as the difference in heatsinks. Can you set the system up perfectly? Can you do it several times in a row? And how closely can you build the reader to the Maxim specs? Can you compare numbers with other people? Probably not unless they all get their diode reader from the same manufacturer. And have the same lead wires. Of the same length. Does the kind of solder on the socket pins matter? This is getting fun...
Here is something more troubling: my experience is that the diode temperature is lower by 4-5C than the temperature of a DOW probe epoxied under the core of a ceramic Duron "Morgan". How can this be that the temperature is hotter away from the source than at the source? It isn't exclusive to my reader: Joe C's $150 maxim test board shows the same phenomenon. Why? I am still thinking about that one.
And finally, to really complicate things, while the diode reader isn't good enough for serious testing, it is good enough to pick up on inconsistencies in methodology. Apply the thermal paste differently? Use more pressure on one of the replicates? All that will show up. And since the response time is very rapid and the resolution of the readers is only 1C (unless you hack the SMBus drivers for lower sampling rate) then the effect is marked. Is this a good thing? The effects after all were always there we just couldn't see them before. Can we come up with truly consistent paste application? Maybe Joe and Bill have thoughts...
Having explained why the diode readers are not suitable for high end testing, I still think that they are far superior to ANY of the alternatives available to overclockers. It picks up on temp changes from things like dragging windows, a bad application of thermal paste, a kink in a hose, and many other variables that the in socket thermiostor or a compunurse cannot. Its temperature compression is also quite a bit less than either a thermistor on the side of the core or the insocket probe. And since it connects to SMBus, you can easily add in a relay connected to the OverTMP function on the reader and turn off power to the computer (PWR_GND) if the CPU gets above a certain point.
As far as variability in the diodes of chips themselves, I haven't ovbserved that. I have tested out I guess 4 different XP CPUs (3 1600+ and a 1900+) and at 1400 MHz and 1600MHz (1.85V) they all produced similar delta Ts (within 1C) when using the same heatsink. Now that is just anecdotal evidence mind you but it certainly is more consistent than you will get when comparing different CPUs with a compunurse that you have to pull off and reattach.
I don't understand how Joe, Joe C, BillA, and other serious testers keep at it. Bill's adage is true:
"All temps are crap...."
And yet it is the job of a few unfortunates to try and make sense of them. The choice of some hardware sites to review neon lights instead starts to make sense. Maybe it is best to simply evaluate cooling gear based upon how far you can throw it?
|