ebastler's low current battery tests

[ebastler (Moritz)]

Introduction

I recently got my hands on a couple of LG INR21700-M58T, and since nobody tested those unicorn-batteries before, I figured I’d grab my old RC charger (Junsi 106b+), check whether it is accurate with my Fluke DMM (turns out, it is) and have it log some discharge curves. And since I was already at it, I decided to continue with other cells I had at hand.

My test setup is limited to approx. 4A with LiIon cells (20W max), so I could only test lower loads - but to be honest, lower loads are what matters for me, personally. I am a fan of sustained performance in flashlights, and short super bright turbo boosts do not really matter much for me. So here we are, 21700 cells tested at 4A, 2A, 1A and 0.5A.

Test setup and methodology

As mentioned, I am using a Junsi 106b+ as a charger/datalogger. It makes use of 4 wire measurement, where the cell voltage is determined through 2 separate leads connected to the balancer port, instead of using the main wires for higher accuracy. These are soldered to the battery holder as close to the cell contact as possible. Sadly I can not contact the cell directly, so expect some voltage drop to still be present across the cell-to-holder contacts.

The cells are charged to 4.2 V, then the charger is set to wait for 10 minutes before beginning discharge down to 3.2 V.

The Junsi takes a timestamped datapoint for cell voltage and current every 2 seconds. These are logged, and then used to compute capacity and energy in a second step, when the graphs are drawn with R.

Results

LG INR21700-M58T

Samsung INR21700-50S

EVE INR21700-50E

Sofirn 21700 flat top (5000 mAh, from LT1s)

Conclusion

I hope these measurements have any value for the community. I do not have the high quality measuring equipment or access to as many cells as others, for example Mooch (I picked 3.2 V as an end of discharge, so my energy results would be comparable to Mooch’s “E-score”). I just had fun measuring, and decided to just put my results out there. Maybe they are interesting to someone.

If you have any comments or suggestions on how to improve, go ahead. My main uncertainty is, whether or not I should discharge the cells lower, as Anduril seems to enter LVP only around 3.0 V, or stick to my current 3.2 V end of discharge. Are there any metrics I am currently not publishing but you’d be interested in seeing?

Scheduled cells

This section lists cells I have access to, and plan to measure in the upcoming days/weeks. I will update the start post accoringly when a new cell drops.

• EVE INR21700-58E
• Ampace JP40

Not scheduled cells

This section lists cells I have access to, but no current plans to measure (for example, due to a lack of a matching cell holder). If people mention that they are interested in a particular cell, I can try and figure out a way to add it to the queue.

• Keepower UH1835P (1200 mAh 18350)
• Vapcell M11v2 (1100 mAh 18350)
• Unknown Sofirn 18350 1100 mAh (included with SC13)
• Unknown Lumintop 14500 920 mAh with USB-C
• Unknown Wurkkos 14500 900 mAh (included with HD10, possibly TS10v2)
• DLG INR14500-095

[gravelmonkey]

Glad the overnight discharge test went well, I was rather worried I’d see a thread “I burned my house down”!

The graphs look good, thanks for taking the time to do these, I’m looking forward to the Sofirn test.

[Mandrake50]

Very interesting! Thanks for doing the work to get these results!!

[ebastler (Moritz)]

EVE INR21700-50E added. Seems like a good value cell, it performs great for low current applications. You can see that the voltage sags a lot more between 0.5A and 4A already compared to the 50S and even the M58T however, so it will likely perform significantly worse at higher currents.

Also interesting how the 50E drops off sharply towards the end of the measurement range (esp. in 0.5 and 1A tests), the 50S a bit less steeply, and the M58T stays way flatter. This tells me that if I discharged the cells deeper, e.g. to 3 V or 2.8 V, I would not get significantly more capacity out of the EVE, while the Samsung still has some reserves, and the LG could probably deliver quite a lot still. Interesting to see, how different chemistries store different amounts of energy at different cell voltages.

[ebastler (Moritz)]

I’m back from a few days vacation and university stress, just began testing the sofirn cell. Will try if I can test 26650s in my holder, too, as I have a 26650 5500 mAh sofirn cell from a SP33S as well.

EDIT: 4A test is done, not great, but not terrible either. Running 0.5A over night again, 2A tomorrow morning and 1A tomorrow afternoon if all goes according to plan.

[ebastler (Moritz)]

Sofirn 21700 5000mAh flat top added.

Nothing too surprising. The worst performing cell in my test lineup so far, but still decent. EVE 50E is clearly better at any current.

[ebastler (Moritz)]

Got 2 interesting new contenders.

With the JP40 I am mainly interested how well the new gen tabless high current cells can sustain their nominal voltage at low currents. I expect a very flat curve with a steep dropoff.

The EVE 58E will be a very interesting contender for the LG M58T. Same specs. Since the EVE 50E exceeded my expectations a lot, I’m curious about this one.

I also have a F60 here, but sadly my gf leaves tomorrow and it’s her light, so I don’t have the time to measure it.

The new Samsung 50S2 would be very interesting as well. I think I will buy a M21H if Simon launches it in other colors, and get it with one of those.

If you want anything else tested, feel free to send me a cell over I’m not Mooch, but IMO my tests are pretty useful for lower power sustained lights, like S21E or similars.

Note: since I am doing this for fun only, I only have single samples available for most cells. I may have outliers or cells with issues that skew the data.

[zak.wilson (Zak)]

I appreciate the low-current tests as they better reflect flashlight use cases. You could probably go a little higher to simulate sustainable medium modes - perhaps 2A for 21700.

3.2V seems a bit premature, and I would consider it a minor flaw if a light cut off that early. Of course it might step down from a high load and quickly rebound to 3.2V, but it would be unreasonable to shut off from a low mode at 3.2V. Many lights I’ve tested shut off around 2.7V.

[ebastler (Moritz)]

I think I will re-run the tests for all cells with 2.8 indeed. Originally that was an accidental setting, then I didn’t want to re-run, but it just makes more sense to go lower…

4A is the hard limit my charger can do in discharge (20W). No way around that.

In theory I could connect a low value high power resistor in series with the battery and have that one bleed most of the power (the charger supports this), but that’s a hassle. In that case it could go beyond 100W I think.

[zak.wilson (Zak)]

I don’t think a 4A test would add a lot when Mooch is doing 5A tests, assuming you test the same cells. I think a 2A test would be pretty valuable, as that’s about the load of the highest mode that doesn’t thermal throttle in a lot of flashlights.

[ebastler (Moritz)]

I’m doing 0.5 / 1 / 2 / 4 for any cell

If I ever end up measuring smaller cells like 14500 I will probably pick different currents, but for 18650/21700/26650 these are good imo.

The 4A one adds very little effort as it’s pretty short. I just set the charger once, click go, and it runs the cycle in 2 hours. It’s the 0.5A one that’s super annoying as it takes over 10 hours. I usually don’t like to let Li charge/discharge running unattended, but 0.5A and 1A I pretty much have to. Esp if I’m at work.

[ebastler (Moritz)]

Oh, if I re-measure - should I use one of my older or newer M58T? They were all bought the same time, some have 1-2 cycles, some 10-20. I’d rather grab one of the almost new ones, same for the EVE 50E. Samsung 50S1 both have a bunch of cycles…

[Hoop]

@zak.wilson

I don’t think a 4A test would add a lot when Mooch is doing 5A tests, assuming you test the same cells.

The more testing by more people the better, I think.

[torchythebatteryboy]

Great work. The Junsi machines are really good. I used the 106B+ when I first got into testing. I was only interested in finding out if Li-Ion batteries were anywhere near the claimed capacity (usually not) and discharged down to 2.5v at 1A as there is hardly any difference in the result at 1A or 0.5A.

[ebastler (Moritz)]

I got my first ages ago back in the days of NiMh RC cars, then later got a second during my race drone days (after I left the first to my dad), and a third after a friend broke his and managed to repair it

Very satisfied with those things. Factory calib of 2 out of 3 is excellent, third is acceptable (always compared to a Fluke 177). For these tests I picked the one where current and cell 1 voltage were closest to my Fluke readings.

[ebastler (Moritz)]

Getting ready for new measurement series

[mcjtom]

What does 20 1S CYCLE 1 mean in this setup?

[ebastler (Moritz)]

4.20V charge, single cell (charger can do up to 6S packs), it is set to cycle mode and only 1 cycle (aka full charge, wait, full discharge, finish).

To be precise, it’s set up to charge to 4.2V, wait for 10min then discharge at 4A until 2.8V and stop.

This is what the software records (Ampace JP40, 4A). I do all 4 measurements (4A, 2A, 1A, 0.5A) in one go, then save the files as CSV for evaluation in my own R script that generates the plots I have posted in the start post. Now I am re-measuring all my cells down to 2.8V.

[mcjtom]

Who else uses termination voltage of 2.8V? Mooch I understand has been doing his to 3.2V. The nominal capacity at 0.2C? is measured to 2.5V? Each can be useful if interpreted well, given the sag and LVP, but I wish there was a single standard cutoff. I kind of like 3.2V one because it’s most conservative and may be a realistic estimate of what a battery can do in a regulated light at higher drains, as Mooch explained lately.

[ebastler (Moritz)]

Flashlights usually enter UVP around 2.8V.