(a prepared reply!)
Nikhsub1: A chiller is going to be expensive, at least that's my prespective on it. Jaydee116 hinted at $8'000, and I think that most chillers are going to run in a similar range. You may still be right though.
The reason that I push this point, is because I look at it from the perspective of the WBTA, which is to set testing standards, and not so much to impose the type of equipment needed, to some extent.
So if we can define those parameters, then that leaves the test bench builder the option of going with whatever solution he sees fit, as long as it meets the standard.
Quote:
|
Isn't the theory that if you want within .1C accuracy you need a measurement increment of .01C?
|
All I know for sure, is that all the errors are cumulative, so the less we have at each point, the more accurate the end result is going to be. We can take this one up in the accuracy thread <here>
Winewood: yes, you right, but the temperature still needs to be maintained.
To recap:
We went from "it must be a chiller because it's perfect" to "the temp needs to remain rock solid".
A chiller may still induce temperature fluctuations. So the question became: how much of a fluctuation is acceptable?
Then I suggested that the acceptable level of fluctuation is going to depend on how a block would respond to it. So I pointed out that if we knew how fast a Cascade responded to a temperature fluctuation, we could keep the fluctuation under that, and set the standard.
Then I suggested that those fluctuations may be more dramatic on a heat die, because of its nature (being possibly copper, instead of silicone). In other words, even if a CPU didn't see the fluctuation, because it went by so fast, a copper heat die might still catch it.
Moving along...
There's two things I have to mention here:
1-It's possible, and I've thought of this, that we may need some ultra accurate measurement tools, not for testing, but for building the test bench.
2-There might have to be an acceptable level of temp fluctuations, as measured at the heat die. Then again, it might be possible to keep it at zero fluctuation, even if the coolant temp fluctuates by some temperature range, over an X period of time. If we target zero fluctuations, which I think (off hand) may be possible, as measured at the heat die, then we 'd have to define the timing or frequency of those temperature measurements.
I feel like I'm just running myself in a circle here. The answer may be absurdly simple: keep the die temp steady.
The problem is that if I only measure the die temp say, every second, I may be missing a big fluctuation. So I guess what I'm eventually asking is, how often (what frequency) do we measure the die temp?
So I'm back to: it depends on how much of a fluctuation you expect from the block, which in turn, leads back to how much of a fluctuation the cooling solution is going to induce.
Aargh!
Ok, let's try this: the magnitude of the fluctuation of the temp measurement at the heat die, has to be zero. In order to make sure that it is indeed zero, we need to sample that temp within time interval "X". Time interval "X" depends on the block, and the cooling solution.
Given that the heat die temp probe will return a result of say 35.7 deg C, that the room temp is a steady 25.0 deg C, all with an error margin of +/- 0.1, then the coolant temp has to remain between 24.95 and 25.05,
on average.
(I'll think about it some more. As pHaestus said: it's not trivial)
If we use a rad/fan combo as the cooling solution, where the speed of the fan is controlled by a simple rheostat which is set once so that the coolant is at 25.0 deg C, then the room temp cannot fluctuate, otherwise the coolant temp is going to fluctuate too.
Unfortunately, most AC or a furnace react within a margin of 2 or 3 deg C (that's so the units don't come on and off constantly). In short, the ambient temp can be expected to fluctuate. This is why a chiller is really handy, because it handles that very easily.
In a rad/fan combo, there needs to be an additional circuit that controls the fan speed, to compensate for that temperature swing. IF it's driven by the water temp alone, then the circuit is going to allow some fluctuation, in order to react. If it's based on the air temp, then it can hold a temperature steady, given that the circuit is designed in such a way that it knows exactly how much faster a fan must turn, but it still must know the water temp, as part of that scheme, and it cannot allow the temp to fluctuate by more than 0.05 deg C, which is half the error margin of the heat die temp probe.
Possible, but not easy.
More ranting later.