XP-L2 V5 Output & Death test by Texas_Ace Over 2200 lumens! Still worked after 15 Amps!

Flashlight Modding and DIY Parts
145

The tests you posted are not directly comparable due to different bins, tints and sources. Although even then if you factor in the Cree binning margin for error into the mix, it is still around 5% for those tests. Don’t forget that each bin can vary by as much as 14%.

I have tested the “same” LED’s on my own sphere purchased at different times from different places and got wildly different results.

Cree builds in a 14% margin for error into each LED bin. So each LED can vary as much as 14% and still be binned the same.

The highest single XP-L HI I have seen reported is around 1400-1500 lumens by many different people. With the new V3 there might be one a bit above 1500 now days. Yet those numbers say that we should be seeing almost 1800 lumens @ 6A with a U5.

Sorry, that just won’t happen in the real world, If it would why does the Astrolux S2 only make about 1300 lumens @ ~6A with a V2 emitter? Interestingly, that is exactly what my test says it should make with reflector and lens losses….

Now don’t get me wrong, Djozz’s tests are great and his latest setup is much improved but these numbers in particular just don’t correlate with real world results. I won’t take a guess as to what caused the high numbers.

The only directly comparable test me and Djozz have publicly posted are the 3rd gen square test where we both tested LED’s from the same reel:


1560 / 1480 = 1.054 Or 5.4% difference in output.

We also compared a few other LED’s via PM and we consistently came up with a ~5% difference.

Put simply the only way to compare these between setups is by testing with the exact same LED’s from the same reel.

We have done this and we have talked about it and ~5% is what the comparable tests came out to.

This is not an issue of weather our tests match up, that doesn’t really matter. What matters is that so far everyone that has built a light using the LED’s in my tests has posted results right in line with the results I got. So thus these tests correlate with the real world results people can expect. Which is the whole point of the tests.

This tells me that my hunch of not running with thermal paste would make the tests more real world accurate for flashlight use was in fact true and a resounding success.

I could not be happier with the outcome.

146

Where the heck did you get 14% from? Source?
There would be no point in binning them into 20 lumen increments if they varied by 50+ lumens…

147

Each bin is 7% higher then the last. This inherently means that each bin can vary by 7% above and 7% below the rated output. Add that up and you get a total of 14% of error margin between bins.

I have seen every bit of this amount of error when buying the same tint and bin LED from different suppliers or even just different times / reels.

This is why the only comparisons between my tests and djozz that matter are the ones that came from the same reels. Which the Square did and it is 5.4% difference in output.

Since the LED’s were identical that means that our test setups were the only variable. Being 5% off is amazing and more luck then anything to be honest. We both were expecting 10-20%.

148

No…
The bins do not increase by percentage, they increase by 20 lumens.
Bin, min, max
T3 220 240
T4 240 260
T5 260 280
T6 280 300
U2 300 320
U3 320 340
U4 340 360
U5 360 380
U6 380 400
V2 400 420
V3 420 440
V4 440 460
V5 460 480
V6 480 500
W2 500 520
W3 520 540

For T3, 20 lumens / 220 lumens = 10%
For W3, 20 lumens / 520 lumens = 3.8%

The bins are not separated by percentage otherwise the % would be constant.
The number which is constant is the lumen delta.

149

Actually that depends a little on the LED that you consider. I just checked for XP-l and XM-L2. There a bin increase of 1 corresponds to 20 more lumens ( at 85°C and 1050mA). For the XP-G2 an increase of 1 bin corresponds to 8 or 9 more lumen (at 85°C and 350mA). However each datasheets states:
“Cree maintains a tolerance of ±7% on flux and power measurements, ±0.005 on chromaticity (CCx, CCy) measurements and a
tolerance of ±2 on CRI measurements.”

150

You forgot that 20 Lumen difference at binning Current are much more at maximum manufacturer Current, and very more at 10A or 15A.
For exampe with 10 times more current and 7/8 times more Lumenens, there are more difference than 20Lumens between two binnings.

151

Someone is arguing for the sake of arguing here no names mentioned lol...

Besides thermal paste is for pussies real men solder to pill ;)

10a is about the limit for these which is amazing really considering the size the light output is immense i am suprised the bond wires didnt melt before the dome started lifting. My power supply only goes to 10a so i couldnt finish the job unfortunately

152

Soldering is the best option but hard to do without getting air bubbles stuck underneath, and also makes upgrading a pain because you can’t just remove two screws and swap MCPBC+led.

153

Looking forward to see your test results and of course the setup Enderman :slight_smile:

It will be valuable and educational for us all to see different approaches. There are much more lurkers than doers here. I’m sure many will appreciate what we can see and compare than just talks.
+5 to all the doers
+1 to all the speakers
+10 to all the doers that share their findings
–10 to all those ……. (fill the blanks anyone?)

- Clemence

154

Yeah big thanks to all the doers too. I mean i do quick tests but i would never go the extemes that some do but its interesting to compare and the graphs are useful!

155

Thermal paste just avoids hotspots. If the surface is nice an there is already a heatspreader in between, the benefit diminishes as a hotspot between spreader and sink is not such a big problem in our application. If the hotspot were directly on the heat source the problem would be real. Luckily the LEDs are soldered to the board.

It surely will not create a difference of LED burning up and LED staying cool.

its still free lumens though, thus I always use thermal paste.

156

Yeah, some day I plan to get myself the proper gear for doing controlled testing myself.
I already have good voltage, current and lux measurement, I still need an integration sphere though.
Maybe one like this?

I won’t do these kinds of “flashlight-host-like” tests, I actually want to see how well the LEDs can perform in an ideal scenario with the best cooling possible, with controlled ambient temperature, etc etc etc

You can already see from djozz’s tests that with a tiny bit of thermal paste you get hundreds more lumens, just imagine what a decent watercooling setup and liquid metal paste can do :slight_smile:

157

This is not true, when all else is kept the same, the difference is ~5.4% as seen in a direct comparison test with the only variable being lack of thermal paste and my setup vs his.

This is about the same I noted in my own tests, 3-5% difference with paste properly applied.

I was mistaken though, it is a 7% margin for error not a 14% with Cree LED’s. Although that is only guaranteed within the rated current range, that goes out the window when you are driving them with 2-4X the rated current.

It also does not factor in dishonest sellers mis-marking LED’s causing even larger variances.

158

I already proved you wrong about this a page ago, go back and look at the your vs djozz’s graphs.

1830 - 1580 = 250 lumens.
250 / 1830 = 13.7% difference, not 5.4%

2) This is a U5 vs V2 bin too, and U5 is TWO bins below V2. Yet the lower bin is hundreds of lumens brighter? lol

3) Then there’s the fact that part of djozz’s LED failed and that’s why the curve shifts down at 6A
If that part of the LED had not failed you can tell that it would be at 1900+ just by looking at the graph.

4) same thing with the XP-G3, his results are hundreds of lumens higher than yours.
Seriously, this is ridiculous, why are you arguing against your own data?

159

This will be my last response on this matter. It is a long read but If people read this entire post, I think people will have enough data to make up their own minds.

Those tests are not comparable as I showed you before.

The tests that ARE comparable are all that matter. And they show a 5% difference.

The fact is that a U5 XP-L HI making 1900 lumens is simply not possible. It doesn’t line up with a single trusted members real world results. It is up to you to show that those results are valid and comparable. They are not.

If they were representative of the real world results then you would see more then ~1400-1500 lumens out of a XP-L HI V2 (which lines up perfectly with what my results says it should do BTW)

Here are some paper numbers for you.

Compare the results of that xp-l HI test to official Cree specs: http://pct.cree.com/dt/index.html

The XP-L HI U5 should be putting out ~900 lumens at 3A with a 50C temp, yet in the test you are saying is comparable to mine it puts out over 1100 lumens at 3A. A difference of 22% from official cree specs

Even Cree disagrees with your claims.

Now lets compare my results to the official cree specs:

My XP-L HI V2 is reading 1005 lumens at 3A compared to the official cree sheet which says it should be 991 lumens… A difference of 1.4%

If you go run my other tests you will see that they also line up within the margin of error on the LED’s from trusted sources for the most part.

If you factor in the difference from the official cree specs of 22% then you end up with 1900 lumens - 22% =1482 lumens + 46 lumens for the bin difference = 1528 lumens You then have a 7% margin for error on top of this, bringing the number up to a max of 1630 lumens

Now if we run my numbers in the same way we have a 1.4% difference from the spec sheet. 1620 lumens - 1.4% = 1597 with a 7% margin of error bringing the number down to as low as 1485 lumens.

So as you can see the numbers overlap great when you look at the data from an unbiased perspective. The tests are not comparable though as you can obviously tell because there are too many margins of error involved to get results remotely comparable.

The XP-G3 tests tell a similar story, his numbers are 7% high and my numbers are 6% low compared to the official cree specs. So figure that he got a high binned emitter and I got a low binned emitter.

When you factor in these differences in the XP-G3 numbers you end up at 1636 lumens for his XP-G3 test. Mine ends up at 1668 lumens. That is a 1.9% difference between our numbers.

Yet even with those big variances to deal with, when you run all the numbers, you still end up with less then 5% margin of error between our numbers. Way better then either of us thought it would be. The same goes for other tests I have done on the “same” LED’s purchased from different suppliers. Large swings in output, Cree uses the 7% margin of error liberally IMHO and sellers are not always honest about what you are getting.

This is why these tests are not comparable, you can’t compare a single part of the system with these tests, you have to factor in a bunch of different factors and that is not science. With margins of error this large you can make the paper numbers say whatever you want

In the test where we ran the same exact LED from the same reel and the only variation was our setups and thermal paste. The results were 5.4% different This IS comparable as there are virtually no margins of error with the LED itself to factor in.

Just because I don’t put a lot of stock in paper numbers doesn’t mean I don’t use them. I like to do all the math on paper and see what something should do, then I go out and test it in the real world to see if things match up.

After I see what it does in the real world, I toss out the paper numbers and focus on whatever I find to be true in the real world.

Wen I say I ran extensive tests when setting up my test protocol, I ran months worth of tests to get the best numbers I could possibly get that lined up as good as possible to real world results.

If you disagree with the test setup and results that is fine, it is a free country and you are welcome to that opinion.

I on the other hand could not be happier with the results. They line up with the cree data sheets. The line up with real world results and use. They give people a good indication of what they can expect from a given LED in the real world. Exactly what I wanted since this is a flashlight board.

Even a mainstream lighting setup would not run them any cooler then my heat sink setup, cooling costs money and noise, things that lighting manufactures are not willing to sacrifice anymore then they positively must. Generally lighting LED’s run much hotter then I am.

I am starting to understand why Djozz got so burned out running tests though, I don’t see me running too many more with my life getting very busy going forward anyways.

I even tested all of this for you, real world testing shows that the difference between thermal paste and no thermal paste with high clamping force is less then 1% or about 15 lumens. Not even remotely the hundreds of lumens you claimed.

Just how important is thermal paste? The results may surprise some people... Tested by Texas_Ace

This is why I don’t put much weight on paper numbers.

160

P.S. Djozz and everyone else, please do not take this as in anyway putting down djozz’s tests. As our direct comparison tests show, our setups read very similarly, so I can’t say much anyways.

He puts in a lot of hard work and will hopefully keep doing so. I know I will be using his data going forward anytime I want data for a new LED on the market.

:+1:

P.P.S. That said Those particular XP-L HI numbers are off for some reason, why I have no idea but I have had tests with numbers that just don’t make sense as well. It just happens. All sorts of possible reasons. It is an old test with his old test setup IIRC so not worth talking about.

161

So your point is that since there is such large variation, neither your tests nor djozz’s test mean anything because they never represent the real numbers?

Half your post is saying “there is huge variation” and the other half is “they line up with cree datasheets and real world results”, which is it? xD

You also contradicted yourself when saying that the temps were fine and you could “touch the PCB” and at the same time saying the dome is burning off.

…

At the very least your tests are comparable between eachother… but that’s about it.
The difference between flashlight host size also means that your results are only representing one scenario, A large SR90 sized host provides much better cooling than an S2+ host, which is what your “no-thermal-paste” setup seems to replicate.
Since temperatures aren’t included at each point in the graph then it’s not possible replicate these numbers at all, which is one of the main requirements of the scientific process.
With temperatures, it would be possible to replicate the same scenario because at the same voltage, current and temperature the only variable that you are testing is the bin itself.

If you actually ran months of tests then I would have expected a proper statistical graph with error bars, standard deviation, multiple samples, an error quantization histogram, and die temperatures :confused:

PS - no I don’t believe djozz got a magically good U5 and you got a horrible V2.

162

Thank you all for doing such LED tests. It is very helpfull and I know it takes much time.

163

+1!!!
Thanks a ton, TA and djozz, for all the tests you provide, your time and effort.

164

2019, I am looking to build a new work light with 10 hour runtime. So, 2 cell, 1 to 2 watts max.

Obviously, the buckpuck 700 would have improved runtime, with lower fV over xpl1. Cri improved over v6 4k xpl1.

However, I can make it modular by using the amc7135. The rear cell can be removed, for a single cell edc with typical pathetic runtime short of a day at a work level of lux needed to see extreme detail 2 meters and down. This design allows for safer charging, and ability to have a pwm dimming board on rear battery pack, for easiest, best control of current with dial, like the buckpuck. Hey, I use my lights for work. Runtime is everything.

So, anyone confirmed if the amc7135 will make use of the lower forward voltage for better runtime? Or does it just burn off the excess voltage? I never use cells not rated down to 2.5 volts. My buckpuck cut out just before this. The amc7135 cuts out somewhere above 2.93 V. I can only access 85%rated cell capacity with the amc7135, while I get 100% capacity out of the buckpuck 700.

And yes, I exclusively do white wall inspection. Hoping eye rods are not off kilter by spill tint shift. I see no beam shot indoor yet. Higher cri, with no discernable drop in intensity is the goal, along with increasing runtime.