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

I’ve mounted many CPU:s during the years and learned a thing or two:

- Never reuse thermal paste

- Clean surfaces carefully before applying

- Apply thin and even

- Avoid bubbles at all costs (a small blob in the center that expands when the mounting force is applied often minimizes the risk of bubbles)

- Make sure that the force of the cooler’s mounting springs is high enough

  • Thermal paste is absolutely essential

To be fair CPUs have something like 125w TDP, XPL2 is like 10 watts lol.

True, but the principle is the same, just even more critical.

Only 8 degrees difference between the very best-rated thermal compound, and lipstick toothpaste or diaper ointment. I don’t think the brand of thermal compound used matters that much.

Thank you.
Facts are highly appreciated.
Great work.

I’d like to see an added comparison, with the MCPCB soldered to the pill.
Solder has always seemed like the best thermal transfer material, to me.

Do note that this table itself does not say anything about boundary conditions, i.e. surface properties like roughness and presence of radius of curvature. It is interesting to see the difference between an arbitrary compound and not having any thermal compound (second from bottom); roughly 35 vs 62, which is quite a difference.

Think about how entertaining your flashlight will be when you tell your friends it was assembled with used diaper ointment! :sunglasses:

Exactly…. this would be the ointment of choice as far as I am concerned…… :wink:

Aww…… :weary:
You did the test first!
Great job BTW :+1:

So Tex, our gut feeling was telling the truth. Not much different results from not usng any paste.
I still prefer the least amount of my trusted moly-graphite bearing grease and a fine lapped MCPCB (and base/heatsink whenever possible).

- Clemence

But not many people realize the big picture. They get bullied by the upper rank brands. Good for marketing :stuck_out_tongue:

When you enter the realm of crazy high TDP and cooling solutions sucking the heat away like beeeeeeeeeeeeeeeeep, thermal paste differences are way more noticable.

But for our flashlight uses, it’s a moot point.

IMO, that chart is misleading at best, for us.
Its for cpu/heatsinks… not exactly the same parameters for flashlights.

A properly cooled and assembled cpu might have what ? a .0001” (or less) of TP film between it and its heat sink. Where as most of our off the shelf flashlights are no where near that thin a tp film, of course depending on who put the light together and if it was ‘worked’. Many of us never even check star boards/shelfs/pills for flatness in our modded lights… TP has a bigger job then.

I have commonly gotten copper/alum stars that have a potato chip warp giving many thousands of an inch gaps in the thermal path. Without correcting those gaps(flatness) in these stars/shelfs, I have reservations that .003”(or more) of lipstick filling those voids will transfer heat the same as AS5.

For us, in the flashlite hobby, star boards and their shelf, are not always made perfectly flat like cpu and cpu heat sinks, that are machined flat and smooth. And many times flashlights don’t have similar clamping force as cpu heatsinks. Commonly cpu’s have very very thin film of thermal paste in contact areas. And thin thin or no voids filled with anything (toothpaste.?) is a lot less critical then a gap 100 times or more that size, that must be filled with tp in many lights…

The less expensive TP do work, but tend to be thin and runny vs the better TP that are thick and have better heat transfer properties. The better TP being thicker are harder to apply correctly, but in our tests in the process cooling industry the better TP’s do work better.

That being said, if a star and shelf are perfectly flat and smooth (like a cpu&its heat sink) even with no TP but clamped tight, there will be good heat transfer with a solid path that TP might only help minusculely in those situations in our flashlight application.

Hence the 8° difference in your above TP CPU/heatsink chart. So the job of TP is to fill voids and still create good heat transfer thru the voids. If the voids are small or non-existent then tp has very little to do. If the voids are more substantial like in many of our lights, tp has a bigger job to do that a good quality thick tp probably will do better than diaper ointment…


Got a min online, a few answers.

The excessive thermal paste test was done to represent how I have opened up many many lights and lots of thermal paste suck under the mcpcb. Only very minimal effort was put into seating the mcpcb. A well seated mcpcb can be seen in the other tests.

The mcpcb was not sanded in any shape or form, just a normal cheap DTP copper mcpcb. The surface was far from perfect, I wanted the test to be as close as possible to real world results in a light.

Remember that the LED itself is passing all of the heat through the tiny little 3535 thermal pad with solder and nearly no thermal resistance.

It then has the entire copper mcpcb to pass that heat onto something else, so it doesn’t take anything close to perfect heat transfer to have good results with a dtp mcpcb.

Thermal paste is good for sure, it helps imperfect setups get almost perfect results. This is why I use it in flashlights.

Past that, don’t overthink it, it really won’t make much of a difference as long as you have some level of clamping force, a thin layer of thermal paste and place for the heat to go (aka, a good thermal path to the outside of the light).

For example, divots into the LED shelf do not bother me at all, they will not make a remotely noticeable difference in actual performance at all. Although even a small bump on the same shelf will make a BIG difference.

Now on my test rig with a way overdone heat sink arrangement on the other hand, I don’t bother with thermal paste as it is messy and offers a mere ~1% difference in output, not nearly worth the hassle. Plus I am constantly swapping LED’s, so it is a whole different ball of wax.

Thx Texas_Ace for this great test! I LOVE the fact that you did the excessive paste as I really wish certain manufacturers would pay special heed to that. Some use it like it's going out of style not realizing it's actually exacerbating the problem and not solving it.

Another difference between cheap and good TP is how well it ages. The cheap stuff often dries up after a year or two and doesn’t transfer much heat after drying up. Good stuff stayes soft and transfers heat even after a few years.

Very good test indeed.

Most of cheap China lights are glued with “holly” fujik which is thermal silicone glue.

Your testbed has a much better solution than a regular flashlight.
If we added some thermal resistance there, the difference would be even smaller, right?

Can you elaborate on the method of your test? How did you exactly measure lumens? What was the duration between each recorded luminous flux? How large and what style heatsink? Got any pics?

I’m surprised by the minimal gains with application of thermal paste. I would expect more. What will a test reveal if both comparisons were made with peak current (which appears to be 6.0-6.25A) and a longer run time to see if the high flux can sustain?

The lumens were measured like all my other LED tests, I have a sphere that I have calibrated with hundreds of lights over the years.

The time between readings is not measured exactly but it is around 1 minute between each .25A reading. By the time the test is complete it has usually been running for ~30 minutes or so, plenty of time to reach an equilibrium. This is also why people will see higher 30 second numbers in lights since they are not heatsoaked at that point.

The heatsink is a stock intel heatsink off a 95W CPU IIRC, pretty basic but works good up to around 100w+.

It is not that surprising, an LED is not nearly as sensitive to heat as a CPU, so the temps themselves may be a fair amount higher but it just doesn’t translate to much change in lumens.

All the tests were done in the same way and took about the same time.

Has anyone else, aside from myself, use IC Diamond Thermal Paste
Been using this for a few years now in my PCs, had outstanding thermal performance, even in my S21A builds. (the H1C driver never thermal throttled and the light got so hot about the body, I could not handle it for long!)
Added on top of the fact that it uses Diamond Dust as the thermal material, the goop itself never dries out and starts performing even better after a couple cycles.