The heat flux, or density - amount of heat load x contact area - is what's being completely ignored here. At low densities the quality of the paste becomes less critical. Down at the ~15 watt range, with a 16mm or 20mm dia contact patch, the paste's quality is irrelevant. If the cheap paste fills the voids, and doesn't dry out over time, it will not get any better than that with better paste. It just is not stressed enough for the on-paper advantage of the better paste to ever show up. Raise that to 200 watts, with a 1cm square CPU die, and the quality of the paste can be the difference between stable operation and failure.
Data is worthless if it is not applied to each situation correctly. Context is everything. I am not saying that emitter junction temperature is irrelevant. I am saying that the parts downstream of the LED's thermal pad, once the MCPCB dielectric is eliminated, can be really really poor and still be good enough to keep the junction temp low and stable - because the contact area between those parts is so large.
On paper, different solder alloys have very different thermal conductivities. As do different solder layer thicknesses. Yet, there is zero change in light output between 'bad' 63/37 PbSn and 'good' SnAgCu solder, or between a thick layer and a thin layer. How can this be, if these things 'really do make a difference'? If that were so, and these details resulted in a meaningful difference in junction temperature, wouldn't the 'good' setup give more light output? Why doesn't that happen?
Socket 1156 Intel I5 960 45nm Package size: 37.5mmX37.5mm
CPU die size: 296mm(2)
CPU to heatsink contact area: +/-30mmX30mm
CPU to heatsink contact area: +/-900mm(2)
CPU power stock: 95Watts
All this adds up to about 9.5mm(2) per Watt through the thermal paste if using the CPU cover to calculate Watts per area.
Or 3mm per Watt through the thermal paste if using CPU die size to calculate Watts per area.
Cree XM-L Package size: 3.75mmX3.75mm
Emitter to MCPCB contact area: 3.75mmX3.75mm
MCPCB to Pill contact area (16mm star): +/-200mm(2)
Emitter power stock: 10Watts
All this adds up to about 20mm(2) per Watt through the thermal paste
Pushing an XP-L to 6 Amps would get it to 10mm per Watt through the thermal paste.
To further discuss…
I am not accounting any heat losses through other means… but it does show that you may be totally right comfychair. But then I think about that huge fan blowing air right on the CPU heatsink cooling it down way faster than the body of a flashlight ever could.
I am not saying that cheap thermal paste is not up to the job, just saying that to my limited knowledge, it could possibly make a difference.
I think people are getting confused on the idea of the difference between pastes at this particular scale and thickness is not enough to see, and confusing it with the idea that “there is no difference at all” between pastes.
Running the LED at 20 or 30 watts doesn't mean you have 20 or 30 watts of waste heat to deal with, LEDs aren't 1.5% efficient like Peltier coolers...
Using even Cree's overly pessimistic "75% rule" makes the picture of what the system has to deal with even easier. I think it's plainly obvious this is the explanation for why there's no difference in light output between a thin brass pill, an average aluminum pill, and a completely over-engineered copper pill. At these power levels this stuff just does not matter.
And... there's one joint in the system we can't modify, the joint between the die & substrate and the conductivity of the substrate itself. I know Cree uses material in the substrate that's as good as they know how to make it with current technology, but it still has to be an electrical insulator first, and then a good heat conductor second. Good electrical insulators are usually poor thermal conductors. And the area between the die & bottom of the substrate is the smallest of all in the thermal path, therefore it's the most critical to overall performance, and there's nothing we can do to get in there and improve it. If the bare die were mounted direct to copper, then the other stuff downstream might become the new bottleneck... but as it is now, it's just not. The bottleneck is right there under the die.
I'm so confused from reading this... I haven't seen or heard of any comparative tests on pills for junction temps or resulting lumens. This thread by pflexpro seems to be the best/closest thing around. It does seem to indicate that junction temp and resulting lumens output are dependent on materials in the thermal path, even after the pill. So, is that test not realistic or am I mis-understanding something?
Fujik is not thermal paste, no matter how carefully it's used! It's a completely different category of stuff. If that were an extremely detailed comparison between solder and an asbestos shim, would it be wise to use that data as a meaningful picture of how well solder performs?
comfy - who is this directed to? I linked the P60 wrap test thread.... Nothing to do with thermal paste whatsoever - it's more about materials - copper foil, alum foil, copper tap, alum tape, etc.
Sorry, we were discussing this one earlier, I assumed it was the same thread.
As for the wrap tests... in my opinion, too many variables I don't feel were taken into consideration.
Much the same as the silicon carbide potting issue. If epoxy + silicon carbide works better than no potting at all, that doesn't tell you anything about the role played by the silicon carbide, since there's no comparison to potting with epoxy alone.
true bout the variables, which is why it seems so difficult to do a really valid test. Seems like the best alum<-> copper comparison is the two tapes, but the thickness's and # of wraps is extrememly important, and if you don't consider that, it's not a tru material comparison. However if Reynolds Wrap is all you have for alum, and that spec'd copper tape is all you got, then it is a valid comparison of what you have on hand.
Just sounds like pflexpro takes the testing methodology serious. For thermal greases, one example would be Michael at OSTS who seemed to go to the extreme in designing his own super grease, and testing the effects with thermal imaging video equipment (I assume pretty pricey stuff). It's a tough call to question the premise, motivation, and his methodologies, specially since he was typically using stock flashlight amps, like with the TN31. I don't think he ever raised amps on those lights he modded for throwing. It's just hard to believe he didn't do his homework to justify all the time, effort and costs. I think you are saying there is nothing proven that his special home brewed concoction of thermal grease is no better than your cheap off the shelf MG Chemical 10 oz. silcone based paste.
Edit: Could be he was doing all this before XM-L2's and DTP copper MCPCB's, so maybe that's why?
Does more light come out the front? I don't care about the temperature, I care about how much light there is. If a change in temperature makes more light come out then it's worth looking at. If it doesn't, it's just irrelevant.
An aluminum Sinkpad claims to be 135 W/mK and copper Sinkpad is 385 W/mK, yet Djozz's test shows only something like 3% difference in output? That's within the margin of error! Shouldn't there be more difference than that, if it matters? Will the quality of the thermal paste have more impact than a 135 W/mK MCPCB vs. 385 W/mK MCPCB? Does a LED put out more light on a 20mm MCPCB than it does at the same current on a 16mm MCPCB? No! Because the area of a 16mm MCPCB is many times greater than what's needed to maintain a stable junction temp, and that shifts the bottleneck back to the spot we can't work on, between the die and the bottom of the substrate.
One interesting thing in his numbers, the Vf doesn't always correspond to the light output (and hence the inferred junction temperature). Higher temp equals lower Vf, so how can a run with the same fixed current but lower Vf (which means it's hotter!) put out more light?
The exact same thing applies to a computer’s CPU.
So you’re saying that in a computer environment, anything after the heatsink doesn’t matter much because the bottleneck it at the CPU to heatsink? So all those fans in the case to exhaust hot air is not needed?
*It’s all a system.
Heat transfers when there is a temperature difference. The bigger the difference, the more is transferred. *
So in a flashlight, if I can make that MCPCB cooler by 10C by using better paste, pill and body design, it will make it easier for the heat to transfer at the bottleneck.
This also explains why we need fans to cool the inside of your computer, why a flashlight can run longer at max on colder days etc.
For that analogy about case fans to be accurate, you'd have to propose some method by which the paste under the star increases the ability of the flashlight body to dump heat into the surrounding atmosphere.
We’re talking about getting the heat away from the emitter.
Keeping the idea of the greater the difference in temp, the greater the temp transfer…
Let’s look at it the other way and see how the cold gets to the emitter…
The cold goes from the body, to the pill, to the paste, to the MCPCB and hits a bottleneck. But at that bottleneck, the cooler the MCPCB, the greater the difference in temp to the emitter which helps keep it cooler.
Edit: Just as the heat has to get out, the cold has to get in.
This is why every single part of the system matters.
and to reply directly to your quote comfychair, the method for the body to dump more heat into the surrounding area is to increase its temperature by having a better thermal path to the emitter. This includes everything between the body and the emitter.
The hotter the body, the more heat is being dumped into the atmosphere.
I’ve read the thread up to the current post (#47), but chose to quote this one. Frankly the discussion up to this point has been more civil than I expected, but it’s still only discussion because nobody has any good data on this. I think it’s clear that both sides have good points - that’s normally where we end up having to do some testing.
One critical thing to understand about comfychair’s perspective is that he doesn’t care about die temperature unless it has an effect on light output. I do not take the same position, so my perspective on thermal matters is certainly going to be at least a little different. All other factors being equal I want die temps to go down. If nothing else it’ll improve lumen maintenance, but I think it’s fair to expect some improved performance like Tom E pointed to in pflexpro’s last graph on the wrap thread. Maybe I misunderstand something about that graph though?
In terms of the quoted post… I disagree about the density of CPU vs LED. We’re talking about ~900 square mm for a modern CPU heatspreader (1156) and running the CPU at around 60-90w (draw, not heat output) for stock speeds or the same for a mild OC. So that’s 10 square mm per watt. The back of a 16mm MCPCB is around 200 square mm and the LED is something like 10-30w (draw, not heat output). So basically the same density the way I’m figuring it. And we still see small differences even with active cooling and low temperatures. I expect to see at least small differences at higher temperatures and with passive or hand-assisted cooling as well.
comfychair also makes a good point in that the most critical joint is the one we can’t modify. The best we can do is to increase the temperature delta from the hot side to the cold side of that joint. The higher the delta, the more heat will move across it - regardless of how bad it is.
I’m not convinced that I’m right. I think it’s clear that there’s no data supporting comfychair’s claim, which doesn’t make him wrong either! Good thermal paste (not the best, just good stuff) is relatively affordable. 10g or 20g tubes of decent stuff is pretty inexpensive and you shouldn’t use much per application. So with the lack of data I just don’t have a problem covering that base. I currently use AS Ceramique because I have a ton of it, but I’ve considered switching to MX-2 because it’s easier to apply correctly and is still affordable.
