I spotted a minor weirdness in the datasheet.
The peak output is not at cool temps. It’s not at 25°C. It’s higher.
Furthermore, the loss from heating up to 85°C is just 3%.
I’m not sure if we should believe this…if yes, we can expect it to have real-life performance quite close to this test.
Uh, what datasheet are you looking at…?
The highest output is at –40C, and that’s because that is where the graph ends.
My bad. I looked at LE CG Q8WP datasheet instead. :person_facepalming:
Here the diagram of the LE UW Q8WP (same sample tested earlier) for forward voltage and light flux depending on temperature Tj (in °C).
I put a lot effort to make the thermal resistance from emitter to heatsink as low as possible to avoid errors.
For the next upcoming emitter tests I will publish also diagrams / measurements like this.
The temperature dependent output is really great and new data you are giving us!
I wonder, would you see the same shape curve if you did the experiment at, say, 5A instead of 0.7A? We are getting into very interesting device physics; potentially separating the effects of high current and high temperature on output.
Maybe, but at the moment I'm not able to measure the temperature dependence at more than 3 W LED power. :(
I'm planning a lot more powerful Cool/heat setup for the future, but it's unclear when it would be ready. At this point I can only say - it's done when it's done...
koef3, could you overlay curves from manufacturer datasheet on that picture?
The data of the manufacturer are valid for 1,400 mA only - but I measured with 700 mA due to design of test setup.
I am not sure if there is a significant difference between these measurements so I would rather not overlay curves on this picture...
I did the test with 1,400 mA now, like the testing conditions in Osram's datasheet, including also overlay of curves from the datasheet.
In the datasheet the values are shown up to 120 °C Tj only, in my diagram up to 150 °C. The change in light flux shown in the datasheet matches very well, and the decrease of Vf is lower than stated in the datasheet. Very nice. :)
Thanks 
It looks like the curves change with current, though just slightly.
Raw data
Osram OSTAR Projection Compact LE UW Q8WP
Amps lm Vf
0,20 84 2,70
0,40 154 2,75
0,60 221 2,79
0,80 282 2,82
1,00 340 2,84
1,20 395 2,87
1,40 451 2,89
1,60 503 2,92
1,80 553 2,94
2,00 602 2,96
2,20 650 2,98
2,40 693 2,99
2,60 740 3,01
2,80 784 3,03
3,00 823 3,05
3,20 864 3,06
3,40 909 3,08
3,60 942 3,10
3,80 980 3,11
4,00 1018 3,13
4,20 1054 3,14
4,40 1090 3,16
4,60 1121 3,17
4,80 1155 3,19
5,00 1187 3,20
5,20 1217 3,22
5,40 1246 3,23
5,60 1278 3,25
5,80 1306 3,26
6,00 1333 3,28
6,20 1359 3,29
6,40 1383 3,30
6,60 1406 3,32
6,80 1431 3,33
7,00 1450 3,35
7,20 1471 3,36
7,40 1491 3,37
7,60 1510 3,39
7,80 1526 3,40
8,00 1540 3,42
8,20 1555 3,43
8,40 1567 3,44
8,60 1578 3,46
8,80 1588 3,47
9,00 1596 3,49
9,20 1602 3,50
9,40 1605 3,52
9,60 1607 3,53
9,80 1606 3,55
10,00 1603 3,57
If you create diagrams / comparisons with this data, I would ask you to publish these charts in this topic and also to specify the source of the data used. 
Raw data of more emitter tests will follow in the next time. Stay tuned!
Osram is a step ahead of Cree:
This graph shows the results of their research in the field of high luminance LEDs. At 4A (2A/mm^2) the efficiency of the Q8WP (same die and solder pad size as the XP-G2) is 32% higher than that of the XP-G2 S4 2B de-domed. As the current increases, the difference becomes even larger.
I think it would be more interesting to draw such chart with Watts instead of Amps.
The way you have it, low Vf LEDs have a natural advantage. If a high Vf and low Vf LEDs scored the same on your chart, the high Vf one would make more lumens and higher surface luminosity. So it would do a better job than the low Vf one.
And Q8WP happens to have much lower Vf than XP-G2.
If you’re that concerned about current (f.e. because you want to use linear drivers), compare lm/A over A.
I use current as the x-axis simply because LEDs are current driven devices. When you use a buck-regulator you will drive the LED with a specific current. Thats when you will see the benefit I’m trying to show here. As soon as a flashlight becomes larger with multiple cells this approach is better in my opinion.
EDIT: also making such a chart is actually more difficult because the LEDs have different wattages at different currents. I would have to use umeasured points on the curves between the measurements points and determine each one individually.
Frankly, I don’t see how a driving method justifies mixing units.
And drawing chart is just as easy with amps or watts - your software already interpolates points. It doesn’t even show where the measurement was. So yes, with watts the actual measurement points won’t be aligned on the X axis. But it won’t affect the curves.
Ok, you’re right about that, but I don’t see a need for it right now.
