DiY LED - Active cooling - CXA3070

Since I am waiting for drivers to get here I have nothing to build so I decided to measure the power and effectiveness of active cooling. I got a CPU heatsink (345grams) and 70mm fan and sanded the surface down to a 600 grit (lazy) and attached a Cree CXA3070 with Prolimatech PK2 thermal paste. Driving it with a Meanwell constant current driver that puts out 1.04A so the COB dissipates 37W. The fan is blowing air directly into the heatsink fins. I used an adjustable power supply for the fan so I could change the speed. I estimated Tj by carefully measuring the change in vF with a Fluke 115 multimeter. Ambient and heatsink temp was measured using an infrared baby food thermometer.



-First test was with no fan at all. This heatsink is not designed for passive cooling and would be way too small of course. Ambient temp was 18C. The heatsink reached 42C-45C, hottest in the middle. Tj reached 93 and was still rising slowly when I gave up waiting. For reference, with a correctly sized heatsink I get Tj of 44C with passive cooling and ambient temp of 21C.

-Second test the fan was running slowly at 5.2V. Heatsink stabilized at 26.5C. Tj stabilized at 42.5C. Fan consumed 1.6W.

-Third test turned the fan up to 7V. Heatsink stabilized at 24C. Tj stabilized at 34C. Fan consumed 1.8W.

-Fourth test turned the fan up to 8V. Tj stabilized at 33C. Fan consumed 2W.


So in summary, for a penalty of 5% efficiency, I was able to decrease the amount of aluminum required from 4.75 LBS to .75 LBS and achieve a ridiculously low Tj of 33C. It ads a bit of complexity and wiring to an already very complex installation and also ads the cost of the fans and power adapters. But on the other hand large aluminum heatsinks are getting very expensive and for a large grow they are actually very heavy, requiring a very sturdy frame to support them safely. I am not going to stop using passive cooling based on this test (because I get some active cooling from the circulation fans), but for those who are building DiY systems active cooling can serve them very well.

BTW the Meanwell driver tested 87-88% efficient and the CXA3070 at 1.04A put out 4900 lumens and was very impressive to see coming off that small heatsink. Lit up my whole backyard and into the woods.

There must be a way to get more efficient power adapters for the fans. If my measurements are correct, the fans are actually dissipating a very small wattage. 8V * .05A = .4W. So the adapter efficiency in that case was only 19% (.4W/2.1W=.19)

I tried an old cell phone adapter at 5V and got similar, very poor efficiency numbers (~20%). Although it did get the job done using only .5W. I was able to get a Tj of 50C. That means only a 1.3% efficiency penalty for using active cooling in that case.

Very interesting stuff supra. I know fans saved my arse..

one thing to keep in mind, if you use a fan on a small heatsink. What happens if the fan breaks down? Leds are going to hit meltdown quickly!

I reused a small thermostat on my light. The thermostat is placed on hottest spot of sink (midship) and thermal glued on. If the sink gets too hot (fan failure) drivers get shutoff until it cools down.

another added complexity for active cooling. Pretty easy to wire in..and you could just sacrifice a few leds to heat overload and let the drivers protections kick in I guess

Primo information here!

Curious if I'll have enough cooling to run the 3070s with the 1400 mA driver. The coolers I got were rated for 70watts, and their fans push a good amount of air.

Remarkable results! Thanks for testing. With red leds that kind of a drop in Tj would a mean a 10% difference in output, with whites a lot less?

anomuumi said:

Remarkable results! Thanks for testing. With red leds that kind of a drop in Tj would a mean a 10% difference in output, with whites a lot less?

Click to expand...

one of my tests...

Osram blue and red Golden Dragon on the same Heatsink
red 1.5W - 2.4*0.64 / 60mA is used by the fan./see thread for schematics
blue 2.2W - 3.1*0.7

30% more Power for the blue means also it must run hotter
Vf plotted over time

Temp.-Range for the red Led
the blue led runs about 8°K hotter(can´t find the pic)

running the fan slower (56 Ohm) gives less noise, and only plus 3°K more heat

relative Output for both


Ambient temp. for all test 22 -23°C

Very informative graphs, thanks guod! I don't know where I got that 10% figure, now that I checked with http://pct.cree.com/dt/index.html I got about 7% and 8% for XPE and XPE-2 reds. CXA3070 @ 1A was less than 3% difference.

What does this meeeeean!?!

Spell it out for us dummies.

It means active cooling is pretty awesome if you don't mind the extra complexity and if we can figure out how to power the fans more efficiently. The fans don't need to be run hard, I got great great results at 8V. The fan was dissipating .4W but the power supply was consuming an extra 1.6W. Doesn't sound like much but it means 12% less bud/watt when you factor it in and scale it up.

Great info, very useful as I am planning to do something similar and haven't seen it done until now. I already use a PC power supply in my rig, so adding 12v fans is no problem and doesn't really up the wattage used much.

I successfully ran 6 CPU fans off a single $10 external HDD power supply which claims >80% efficiency:

Gaius said:

Click to expand...

I need to figure out how to use this brand new Tekpower RS232 voltmeter.

Supra, would you be able to give a basic tutorial for running these fan efficiency tests? I'd love to help out!

it's getting to the point where I'm not sure why anyone would buy an a51 despite their quality when you can do 3 x 3070s, drivers, and CPU heatsink/fans for < 1/2 the cost. Not a knock to a51 at all, but.... who's first to churn out cob-based lights and sell them @ $100 markup per light and still undercut the competition?

You would need an accurate measurement of the input watts which would require a kill-a-watt or ensupra. My kill-a-watt isnt precise enough to measure small devices (maybe the newer ones can?) but the ensupra can read right down to .1W.

To calculate the actual fan dissipation power you need to measure voltage (probes in parallel anywhere in the circuit) and current (probes in series anywhere in the circuit). Then V * A = W.

Once you have dissipation wattage divide that by the input power and you get the overall efficiency.

caretaker you have a good point. I figured the a51 190 3700K to be just under 30% efficient, which is awesome for a commercial lamp, but 30% is best case scenario, we dont know the bin used or Tj.

A diy with a 3000K CXA3070 running at 1A is 39% efficient, a much warmer spectrum and much cheaper. I figure a 3000K spectrum has a better chance of success at flowering without reds than a 3700K does.

So $43 for a COB, $11 for a driver and $10 for a heatsink/fan and you get 37W for $64, call it $2/W total cost.

caretak3r said:

it's getting to the point where I'm not sure why anyone would buy an a51 despite their quality when you can do 3 x 3070s, drivers, and CPU heatsink/fans for < 1/2 the cost. Not a knock to a51 at all, but.... who's first to churn out cob-based lights and sell them @ $100 markup per light and still undercut the competition?

Click to expand...

IMO Area 51s are a bit overrated here at RIU. In fact, I am selling my XGS-190 immediately following completion of my the new 3070 array. The XGS is cool an all, but fairly expensive for the coverage when compared to COB tech. Still better than most out there, but DIY is the future.

So if anyone wants a 2-month old XGS, I'll have it for sale in a few days.

SupraSPL said:

You would need an accurate measurement of the input watts which would require a kill-a-watt or ensupra. My kill-a-watt isnt precise enough to measure small devices (maybe the newer ones can?) but the ensupra can read right down to .1W.

To calculate the actual fan dissipation power you need to measure voltage (probes in parallel anywhere in the circuit) and current (probes in series anywhere in the circuit). Then V * A = W.

Once you have dissipation wattage divide that by the input power and you get the overall efficiency.

Click to expand...

I have a kill-a-watt. I'll start taking some readings once I figure out how to use the volt meter.

Does your kill-a-watt read in the decimal range like .5W for example? Mine only does whole numbers and I have seen other kill-a-watts that do read decimals. May not matter much in this case though.

This might help demonstrate the actual connections needed for the measurements. Since I do a lot of testing I made a set of probes specific for current measurements and another set specific for voltage measurements. I use slide connectors for all my circuits so I can bug in that way which is especially helpful for making current measurements.


The set on the left is for voltage so you can bug in in parallel anywhere in the DC side of the circuit. This can be done very easily with the probes if you have good access to bare metal in the circuit. The set on the right allows you to bug-in in series to measure current, same idea. You can make both measurements at once if you have 2 multimeters.

One other thing I have noticed, the multimeter can add some power consumption (.2W in my case) so you don't want to take your input power measurement from the kill-a-watt while the multimeter is in series measuring current.

SupraSPL said:

Does your kill-a-watt read in the decimal range like .5W for example?...

Click to expand...

I don't remember seeing decimals, but I'll double-check when I get home.

This.

Come up with a simple aluminum rail system for pc heat sinks and you got it!

SupraSPL said:

caretaker you have a good point. I figured the a51 190 3700K to be just under 30% efficient, which is awesome for a commercial lamp, but 30% is best case scenario, we dont know the bin used or Tj.

A diy with a 3000K CXA3070 running at 1A is 39% efficient, a much warmer spectrum and much cheaper. I figure a 3000K spectrum has a better chance of success at flowering without reds than a 3700K does.

So $43 for a COB, $11 for a driver and $10 for a heatsink/fan and you get 37W for $64, call it $2/W total cost.

Click to expand...

Did you know that little dot you do not connect anything too is to measure the temperature. Cree goes over it in the data sheet. They recommend measuring the temperature that way because of all of the color spectrums. My Tc is 50c. What is yours with your lowest Tj?