@DoctorDelta9 for some reason Cree does not include the thermal resistance of the package in their specs. Bridgelux does, but I suspect there is more to the story, I think thermal resistance increases as you increase the heatflow through a constant sized thermal interface. In other words, if thermal resistance were a constant we should be seeing a certain minimum junction temp and temp droop at each current regardless of the cooling power of the heatsink but that has not been the case. That is great news for us. Consequently,
when running soft we can get temp droop figures close to zero, meaning we can use the Tj 25C lumen figures in some cases. Even at 2.47A (96W) and low fan speed I have seen Vero29 temp droop at only 2%. This denotes a Tj of less than 50C which seems suspicious. I penalized the Veros by 4.33% in my spreadsheets which is assuming a Tj of 50C. But it seems in practice they may be performing better. I have not tested the CXA/B 3590 but I will soon. I would estimate 2.5-3% droop because the CXA3070 is 4-4.5% in those conditions.
As far as the paste goes, PK3 is 11.2 W/(m*K) conductivity but I agree with church, the thermal junction is almost inconsequential in our application (as long as the paste is covering the entire junction). The reason I believe that is because I have taken the trouble to sand down heatsink surfaces to perfection and mounted a COB with some of the best paste that is reasonably priced (PK3). I mounted a second COB to a stock heatsink with stock paste and ran both hard (Vero 29 @ 100W). I found no difference in thermal performance between the 2, more great news for us.
@alesh has posited the idea that there could be systemic error in my temp droop test method which is a fair concern, for example maybe there is an "instantaneous" drop in output that is too fast for my equipment to detect. I have confidence in my measurements for these reasons.
-I monitor the Vf of the COB and take pulsed readings. When run at low currents there is almost zero change in the Vf but as I increase the current of the test,
I can see the temp droop as it occurs in the slight and gradual decrease in Vf and I can see how long it takes that Vf change to cool back down to room temp (it takes quite a long time when the difference between that section of heatsink is only very slightly higher than the rest of the sink. Once I see that slight temp increase in the heatsink, I can see how stable it is.
-I use my fastest voltmeter for the testing, a Fluke 115 and the lux meter is very fast as well.
-I have done quite a lot of temp droop testing by now. At first I was confounded by lack of repeatability but eventually I found a method that gets great repeatability. There were tricky things throwing measurements off, certain driver's current output would shift during the pulse so I had to find drivers that were steady. I discovered slight Vf changes would occur when touching the connectors to switch fan power adapters, so I learned to move the fan connectors away from the mulitmeter connections and rebuilt my multimeter leads.