I made a blog post about how to make Chip-on-board (COB) based LED fixtures (here is the link). I included materials list, step-by-step assembly details (maybe too much for most people), information about driver selection, and measurement data using PPFD.

Also I came across a couple of useful youtube videos, so the links are provided at the very end of the post. I started DIY COB LEDS with cheap eBay COBs, but the video provides the reason why these are not worth getting. Actually I think they are not worth using even if they are free.

There are newer generations of COB LEDs (Bridgelux Vero Version 2.0 and Cree CXB series), which came out this summer. Some of them (high efficiency bins) are still not easy to get. Some people are achieving close to 70% radiant efficiency with Cree CXB3590 (only 30% of electricity is wasted as heat, and 70% is converted to light energy). A cheaper way to grow plants under artificial light! :)

I posted several other LED related posts here, but COB LEDs are the most promising (cost effective) DIY project for CP, which requires lots of light (similar to MJ).
Last edited:


War. War never changes.
Staff member
Great! I'd been looking forward to this post. I'll check it out!
Thank you, Mark & Est. I didn't include this info, but for the high light plants, I think a combination of 4x Cree CXB3590 and 1x Meanwell HLG-185H-C1400B (orA) is the sweet spot (~200W). High-bin CXB3590 is still difficult to get, but you can get them by contacting Shenzhen KingBrite Electronics Co., Ltd. - Cree led,Bridgelux led $47.5 ea. The driver is around $60. I don't have any association with the linked sellers. CXB3590 comes in 36V or 72V version. 4x 36V version can be connected in serial. With 72V, you need to connect them in 2 parallel circuits. The wall-plug (radiant) efficiency is close to 50% with this combo (about $300+). It will have better PAR quantum efficiency (micromol/J) than any production grow light fixtures.

A lot of good info about CXB3590 (and other LEDs) in LED and other Lighting | Rollitup. Even though I don't grow their kind of plants, I learn a lot from this forum.
Last edited:


Lotsa blue
Thanks for the additional resource, naoki. Amazing LED creations in there. It is easy to get lost.
Naoki thanks for posting this. I am in process of ordering components and can't wait to start. I will post my results when I get going.
Aerogrower, this may be a bit too late (I haven't checked this forum for a while). I was shopping around and I found a very good price on CXB3590 (the current top of the line of Cree CXB series), and I ordered 4 of these. It is 5000K 70CRI, and $35.26 with minimum order of 1. CXB3590 is still tough to get, and the price is usually $46 (from China direct) or much higher from domestic source. This is cd bin (not the top bin), but you can easily get radiant efficiency >50% with this.
This is 72V version, so the choice of the driver becomes limited. From MeanWell, basically you need to use HLG-C series. I ordered HLG-185H-C1400B to drive 4 CXB3590 (2 parallel circuits with 2 serial CXB per circuit, so each get 0.7A), total of around 200W.

They also have cheap CXA3070, 5000K, 80CRI, AB bin (middle bin) $21.47ea, but minimum order is 2. It is not the most recent CXB, so the efficiency is a bit lower. Since the price is low, I'm going to try cheap ghetto liquid/water cooling of LED. I have a relatively small enclosure (3x2x2') which I want to keep it as cool as possible, so this is an experiment. I'm going to run them really soft (at 500mA for 18W ea), which gives respectable efficiency and lowers heat release.
Last edited:


Lotsa blue
Thanks again, naoki! The improved prices certainly move me closer to possibly doing it myself. Meanwhile I am enjoying and learning from your posts.
Great, Mark! I hope you'll give it a try. It did took quite a while to learn about photo-biology and LEDs, but there are lots of good information floating around (it was sometime difficult to filter out the misinformation at first, though).
Naoki...perfect timing I actually am building the low light model you described in the beginning of your thread. This is my first time so I wanted to start small and make sure I could do it. I have all the components except the driver which apparently is on the proverbial slow boat from China. May not be here for several weeks. I will post here when I have light! Next will be a full fledged grow light, so your timing is right on. Please check in here as often as you can in case I run into trouble lol.
Well did my first build with Naoki kinda holding my hand. I built the 50 watt single COB led. It gives off a very nice pattern of light and has inspired me to build a 200 watt light that will cover most of my grow area. Thank you Naoki couldn't have done it without your help. Has anyone else here started a build? If your on the fence please give it a try and post on here. I will try to post some pics later.


War. War never changes.
Staff member
My LEDs are supposed to arrive today. I'll be assembling pieces over the next month and will start putting things together in the start of 2016! I'm sure I'll be in this thread with questions.

Looking forward to seeing your pictures.
Sounds great!

I was wondering if there are some other heatsink choices. This may be interesting to some people. These Mechatronix heatsinks appear to be predrilled for Zhaga compliance (a standard to mount LEDs). So if you use Ideal LED holders, no need to drill/tap (not difficult, but the most tedious part).

ModuLED Mega 134100 0.67C/W CXB3070 or smaller $19.06
ModuLED Giga 152100 0.53C/W CXB3590 or smaller $27.20
IceLED Ultra 0.25C/W CXB3070 or smaller $14.01

They are available from cdi. Unfortunately, they don't ship with USPS, so it's not a good option for people in Alaska or Hawaii.

The lower value of thermal resistance (C/W) is better. ModuLED is passive, and IceLED is active. I didn't look if there is an option to attach a fan to ModuLED (but I'm sure it can be done). Here is some explanation of how to interpret thermal resistance.

Here is a calculation of COB efficiency:
COB efficiency Spreadsheets | Rollitup

So let's say that you drive CXB3070 AB bin at 1.4amp. You can look at "50c diss watt" (dissipated watt (the watt used by LED) at 50C temperature). About 49.83W. Now you can look at the "50C %" column, which says 48.50 %. This is the radiant efficiency (the proportion of electricity converted to light energy). So the rest (51.50%) is converted to heat. This means 49.83 * 0.515 = 25.66245W is released as heat. So with 0.67C/W heatsink, the LED will be ambient (say 25C) + 0.67 * 25.7W + a little bit more (due to the inefficiency of heat from LED to the heatsink, I couldn't find this thermal resistance value for Cree), which is 42.2C + more for the LED temp. I think many manufacturers target the LED (junction) temp of 85C or so, so this would be decent amount of cooling. I'm not sure if this makes sense (or if I'm wrong) since this heat calculation stuff (thermal resistance stuff) is something I recently learned, and I wouldn't say that I understand it fully.

While I was looking around, I saw something called Synthetic Jet (SynJet), which seems to be interesting, but they are kind of pricey. It's supposed to be more reliable and quieter than fans.
Last edited:
It's too bad Digikey has such a limited selection. Looking through the data sheets, it seems like the 90 CRI parts would produce more useful light (for the plants.) Thanks for all the information, I've been researching LEDs on and off for a long time and this is the best build I've seen.
nimbulan, with availability, KingBrite seems to be the easiest source at this moment. I think aerogrower contacted him.

Also, generally, you lose the total PAR output with higher CRI. This may not be an accurate description, but it is easy to understand in this way. Phosphor-based white LEDs are basically blue LEDs + phosphors. When phosphors are hit by blue, they fluoresce at a longer wavelength (yellow-red). So different CRI and K are generally done by changing mostly the phosphor layers. So to get warmer color (lower K) or higher CRI, they generally apply more phosphors. The light can be lost at the phosphor layers, so some photons get lost with more phosphors.

Info below here is probably too much details for most people, but I'm just showing that the above statement isn't based on "myth".

If you want to see more exact number, here is the calculation done by alesh of RIUp forum (well, I added a bit more calculation and re-implemented it in R by myself). Basically, you digitize the Spectral Power Distribution (SPD) curve, and then you can calculate these conversion factors with R (or a spreadsheet).
##                                      ler      qer  par.qer par.per.lum ppfd.per.fc ypf.per.lum
## Cree.3000K.80CRI               327.26551 4.868716 4.662723  0.01424752   0.1533591  0.01288196
## Cree.3000K.90CRI               276.12267 4.915020 4.511533  0.01633887   0.1758702  0.01458771
## Cree.4000K.80CRI               322.10079 4.678840 4.533168  0.01407376   0.1514886  0.01230506
## Cree.5000K.70CRI               323.42007 4.605715 4.471141  0.01382456   0.1488064  0.01204868
## Vero.3000K.80CRI.v1.2          323.10113 4.879973 4.658829  0.01441910   0.1552059  0.01305418
## Vero.3000K.80CRI.v2            319.73025 4.944016 4.704169  0.01471293   0.1583687  0.01346317
## Vero.4000K.80CRI.v1.2          321.96749 4.730732 4.564988  0.01417841   0.1526152  0.01252857
## Vero.4000K.80CRI.v2            323.62228 4.823261 4.619960  0.01427578   0.1536632  0.01281032
## Vero.5000K.70CRI.v1.2.2        331.47526 4.614143 4.489774  0.01354482   0.1457953  0.01179274
## Vero.Decor                     271.76787 4.977243 4.558739  0.01677439   0.1805580  0.01510056
Explanation of columns:
  • ler: Luminous efficacy of radiation. Basically how much lumen is given per 1 watt of emitted light (not 1 watt of electric power went into the diode). The unit is lumen/W
  • qer: quantum efficacy of radiation. I'm not sure if this is the correct term, but instead of lumen in LER, we are using number of photon flux (micromol/s). The unit is micromol/s/W = micromol/J. In this, I'm not using PAR, and using all emitted light.
  • par.qer: Similar to QER, but now we are counting the light between 400-700nm (PAR). unit is micromol/J
  • par.per.lum: For a given lumen, how much PAR PPF is there. Companies only gives the lumen output for a given current and temperature. You can multiply the lumen number with this factor to get the PAR PPF. The unit is micromol/s/lumen
  • ppfd.per.fc: If you have only footcandle meter, you can use this factor to get the PPFD. The unit is micromol/m^2/s/footcandle
  • ypf.per.lum: Yield photon flux per lumen. This quantity is similar to par.per.lum. Blue light is about 20-30% less efficient than red light in terms of photosynthesis. While each photon between 400-700nm is counted equally in PPF, each photon is weighted by the photosynthetic efficiency in YPF without limiting the wave lengths to 400-700nm. This difference in photosynthetic efficiency is expressed by Relative Quantum Efficiency (RQE) of photosynthesis. I used the RQE obtained by McCree (1972). See the figure in this wikipedia. A lot of people are familiar with the chloroplast absorption spectrum (top of the same wikipedia page), but many people are not aware that absorption is just a part of the story; some of the light is absorbed by inactive pigments, which don't contribute to the photosynthesis, or released as heat before it initiate the photosynthetic reactions. McCree's RQE is a more relevant curve than the absorption spectrums of chlorophylls. The unit is micromol/s/lumen.

For the discussion of different CRI, we can focus on par.per.lum column. Then we get the lumen from the datasheet of CXB3070. The value is for 1.9A (higher than I would use) and junction temperature of 85C (higher than what I would target). Forward voltage is 36V, so 68.4W of electric input (not including the loss of power due to AC/DC conversion at the driver). Here are the minimum flux value for the top bin for each.

3000K 80CRI AD bin: 9000lm = 128.2 micromol/s (PPF) = 115.9 micromol/s (YPF)
3000K 90CRI Z2 bin: 7390lm = 120.7 micromol/s (PPF) = 107.8 micromol/s (YPF)
4000K 80CRI BB bin: 9500lm = 133.7 micromol/s (PPF) = 116.9 micromol/s (YPF)
5000K 70CRI BD bin: 10000lm = 138.2 micromol/s (PPF) = 120.5 micromol/s (YPF)

So with 3000K where I could compare the different CRI, lower CRI is better as expected (in both PPF and YPF). Basically if you go up in CRI, the bin number goes down a couple steps.

Now, if you look at PPF values, you may think that 5000K is better than 3000K 80CRI (about 7.8% advantage). From the action spectrum (RQE) of photosynthesis (by McCree 1972 and Inada 1976), one photon of blue light is 20-30% less efficient in terms of photosynthesis than one photon of red light. Although LEDs with lower K have lower PPF, it contains more red light. So the difference in terms of YPF, which incorporates the difference in photosynthetic efficiency, become smaller among different K (4.0% advantage).

Another point is that the PAR efficiencies of these COBs are pretty high. Assuming that we lose about 10% during the AC/DC conversion, so the actual input electricity in this example would be 68.4/0.9 = 76W. With 4000K, 133.7 micromol/s / 76W = 1.76 micromol/J. The top commercial grow light is around 1.6-1.7 micromol/J for both LED and HPS. See table 3 of this paper. To make the output similar to the commercial grow light, you can use 4-5 of these. So the total cost of <$400 can beat the $1000 top LED fixture.

If we use lower current and better heatsinks (to lower the junction temperature), the efficiency improves. Also if we go with CXB3590, the efficiency goes beyond whatever available grow lights. My newest one is CXB3590 at 50W each, which I think gives >2 micromol/J.
Last edited:


War. War never changes.
Staff member
Est what COB leds are you going with? What drivers etc?

I hopped on the CXB3590 deal that Naoki linked earlier in this thread and went all-in for four! That means I'll be going with the driver from the same post unless I happen to find something else that fits the bill better or more cheaply.

I'll be looking into those Mechatronix heatsinks for sure. Not needing to buy and drill several cpu heatsinks would be great. I'm still looking at alternatives though, so if any of you have suggestions, do let me know!
Est, with CXB3590, it has to be ModuLED Giga, which is quite pricey.

I talked about this earlier with aerogrwer, but many people use Extruded Aluminum Heatsinks from HeatSinkUSA. They have different profile, and will cut it for your custom length. 40" of 5.886" profile width should be enough for 4x CXB3590 to do passive cooling without fans. $70 + shipping etc, which is slightly more expensive than CPU fans (cheaper than ModuLED), but it isn't too bad. With passive cooling, you don't have to worry about the fan breaking down. CPU fans are fairly reliable, but there is a chance of break (so I'm adding a thermal protection now). You could add 1x fan in the middle of the big heat sink if you want to go further with efficiency. A small amount of air flow can decrease the temperature of heatsink quite a bit. It would give fairly dense light, maybe similar to 6-8 bulb version of 4' T5HO. If this is too dense (too much light), you could make it to 2x 20".

Here is a related RIU thread, which I got the information from.
After all the discussion we had I decided to go the active cooling route with the computer type fans. I have access to a high speed milling machine that makes short work of the drilling and tapping. I am going to build a 4" x 4' long frame out of aluminum angle, welded at the corners. The heat sinks will fit inside the frame and be screwed to the frame. That way I have a neat way to secure all the wires and a sturdy mounting surface to hang it up when im done.
I will be ordering the COB leds today and im not sure if im going with the CXB3590's or the CXB3070's. So depending on which way I go will determine which driver I get.
Thanks for keeping up on this thread and thanks for all your help Naoki!