I know of some who have milled out the center pad of a noctogon or sink pad and dropped the remaining star over a post. It can and has been done but not for the masses.
Yeah that’s pretty cool Night. You might have issues with shorting it out from the looks, but it skips a layer to the sink. That will be better heat transfer. In flashlights though, the sink isn’t actively cooled in most cases. I was thinking, if someone made a water cooler that fit into an extension tube below the sink before the batteries. That actively cool the sink with copper tubes and coolant, you could see the benefits in keeping the lumens high. I know they’ve already made a huge light like the X18 but, I’m talking affordable.
Not sure there is enough space to fit a pump and rad in a tube that can fit in your hand. And the exhaust might burn your palms 
Maybe that’s a reason it hasn’t been tried on the cheap.
Then you should know that with different heat loads and different contact areas, it's possible for a joint that's critical in one application (CPU-to-HSF, especially a bare-die CPU with a tiny heat path) to not be critical in another, totally different application.
The LED's tiny thermal pad is similar to the tiny bare-die CPU, in that you have a lot of heat that has to pass through a very small area. Back when all hot-rod CPUs had exposed dies, the quality of the thermal compound and flatness of the parts and clamping force was, indeed, critical. After the switch to CPUs with integral heatspreaders, it barely matters anymore. A CPU with an IHS is similar to the connection between a MCPCB and a pill. The contact area is very large compared to the source of heat, the LED's thermal pad. The larger the contact area, the less critical the quality of the joint.
In a flashlight you cannot measure any difference in light output between a copper star secured with screws and using the most bestest thermal paste in the known universe, a copper star secured with screws and the cheapest white grease you can find at Radio Shack, a copper star glued down with JB Weld, and a copper star soldered to the pill - even though, on paper, the 'quality' of those joints should be vastly different. At the heat levels and surface area involved here, it just does not matter. It does not affect the light that comes out the front, which as far as I remember is what a flashlight is supposed to do. Even in the overclocking world, a 'measurable' difference in core temp between one paste and another paste does not mean the max stable CPU speed will be any different. It must be a significant difference, meaning that it changes the operation in some fundamental way (CPU speed, or light output).
Well I have to disagree with you comfy on the overclocking cpu end. It’s pretty common knowledge among us who’ve overclocked cpus to their limits, the thermal compound does matter quite a bit actually in the longevity of that overclock. If you want longevity, 2 or 3 degrees is quite a bit and a lot of the pastes out there, especially silver pastes- they would dry up and fail over time. While I agree with you to a point about delidding, intel noticed the trend in overclocking their low end chips to beat their higher end and they just couldn’t have that. Intels problem is they are so good almost all their chips are winners off the wafer. They switched from soldering the dies to using a cheaper thermal paste in between the die and sink which is why delidding became popular.
AS others have mentioned it has been done by several members here already.
18650, dbc, djozz, Match, O-L, are just names that comes to mind apart from those who has already showed their handywork.
I did it once and never again! As long as I can have sinkpad/noctigon/those new really big ones from Maxtoch I will drill, cut and sand those any day over this solution.
Just my 0.02$
How much power can you shove through a single XML2 before the bond wires vaporize? Since that's what the picture in the OP shows, let's stick to that. Single XML2. How many watts?
Around 35 or so.
If you had a CPU that had a TDP of 35 watts at its maximum overclocked speed, would the thermal paste make any difference? If you move the discussion to a TDP of 200W, then I'll agree, the paste will make a difference. Because there is more heat that has to be moved across the junction. At the levels involved in flashlight stuff it just does not matter.
Comparing like-for-like, copper star, flat surfaces with proper contact, proper clamping force, you will not see a single lumen more with the latest carbon nanotube/diamond hybrid miracle paste than with plain white silicone based paste. Prove me wrong.
I don’t know about 12 amps to an xml2. I’ve heard some have acheived close to 9, I think.
Who was it here that was driving an emitter under water to very high amps for extended periods of time? I don’t know what he got up to.
Trying to solder the wire to the LED after soldering the LED to the post proved most challenging. A better method would have been to high temp solder the wire to the LED first, then low temp solder the LED to the post. This is not a low cost design-for-assembly option.
Just something to keep in mind.
I have a technique to make that wire connection very simple. That is why I use solid 22AWG wire. It is already attached to the driver and can be bent to hold itself in place. A quick touch with a small tipped iron and a little solder; poof, connected. I do this with all my pills.
The idea is that surface tension of the solder on the stub will center the emitter… that is why I use the term “float”. This will avoid the shorts. The real challenge is removing the heat form the pill in time. I’m thinking water around the heating post once the emitter settles.
Many kudos for getting this to work on an XP-G, Crux! I won’t attempt that
Water cooled; hmmmm J)
Sweet! I can see a coffee warmer product here somewhere!
I’ve been wondering this too. The XM-L has 3 bond wires and the XM-L2 only has 2. I suspect the XM-L has a little margin over the XM-L2 for max current if thermals can be managed.
We already know the failure point for XML2s; XML may be able to run at higher current, but it will put out less light and that's what matters, not how much current you can run it at. If you want max current and don't care about how much light it makes replace the LED with a little square piece of copper, you can run really high current with that setup.
35 watts is nowhere near 12 amps... that would require a LED that did 12 amps at 2.9 volts. Not gonna happen. And 35 watts is probably a bit higher than the actual limit, and less power makes the thermal issues even less critical, not more. You cannot put enough power through a XML2 for the quality of the thermal pastes to be an issue (again, just a reminder that we're doing this properly, so no aluminum stars glued to hollow pills in a plastic light).
We do?
Good argument for considering tests using the higher voltage MT-G2. Hmmmmm
MTG2 doesn't have bond wires to blow like little annoying fuses, but the light output still drops off above a certain point regardless of temperature (same way a CPU will have an upper speed limit no matter how extreme the cooling is - some of them won't go any faster at -30 than they do at +30). MTG2's upper limit is still not that extreme, at least not compared to CPU levels of heat vs. contact area. Above 12-13 amps it makes less light the more you shove through it. 12A x 7.6v = 91W.
Plus, depending on which host it's in the outer surface of the light and its ability to shed heat into the outside world will be the limiting factor, once that limit is reached improving thermal performance of the innards is totally pointless.
I was thinking 36V at 3 amps… . and I fully appreciate the limited return of photons. But pushing those bounds is what we are all about, right?
Who said this had to be a handheld? My boundaries are expanding
Even with the early works of the Q5 and the pushing the limits of the XM-L shows there is significant margin in the Cree designs.
Of course, optics are the next horizon, but maximizing an emitter at the bleeding edge is still a worthwhile challenge.
BTW: I did some looking on Noctigon and thermal conductivity. Do the Noctigons now have a coined heat pad or some other way to make it flush with the + and - pads?
I am not happy about the numbers for tin/lead as a thermal barrier. The early reports of varying gaps is somewhat disturbing. If you go through the math, thermal conductivity is indeed a significant factor as you go through the stack.
Seriously, I am not ignorant of all the ramifications of pushing the limits. However, I am well trained in the scientific method. Why in the world would you try to stifle creativity? I appreciate you sharing what is known… but there is a lot to learn with every new opportunity.
I'm not trying to stifle anything, I'm just telling you that what looks 'bad' on paper is not significant in actual use. You are falling for a type of logical fallacy - attributing too much value to things on paper that do not have an effect in real life.
Along those same lines, the different types of solder have been tested in real-world scenarios, not the world of datasheets and whitepapers, and the thermal conductivity of the solder alloy is not significant. It doesn't make any measurable difference. Folks here have even tested a correctly-manufactured direct thermal MCPCB with all the pads level (for the thinnest possible solder layer) against the knock-off copper MCPCBs that have a dielectric layer under the thermal pad, but with the dielectric cut away and the resulting huge trough filled with a much much thicker sub-optimal solder layer, and found they put out exactly the same amount of light.
Other people have already done this work and it's been checked and verified by others - why are you dismissing actual data in favor of math taken from datasheets?