Cursory Google search suggests this VN version uses a dedomed Cree XM-L2 LED and is driven at ~6.5 Amps.
The Cree is a 3V, 5050 footprint LED, if you wanted to reflow an emitter using the current LED MCPCB. If it’s a standard MCPCB, you could also just order any 3V LED you fancy pre-soldered to the board, so long as you can find a suitable LED gasket for your chosen LED and the reflector.
6.5A is quite a reasonable current level, I’m not sure there are any current 5050 LEDs which can’t handle this (?).
I’d say that 6.5A is quite a modest drive current for modern 5050 LEDs, and would be wasted on an SFT42R, which can handle double that. I would suggest a smaller-die LED that could achieves almost the same output, but throws much better. But if the vn mod involves substantially boosting the current, then the 5050 4mm^2 power class like the SFT42R would be a good match.
I think Convoy’s W5050SQ5 would be an excellent fit, coming in many CCTs. The biggest problem with this LED is its high Vf and tendency to lose regulation with single-cell drivers, a problem exactly solved by your setup!
Alternatively, you can go for SFT40 3000K for high CRI. Also, Convoy’s F45R is essentially an SFT42R that is available in lower CCTs, if 3000-4000K is your preference.
If you’re using koef3’s numbers, it’s actually 1850lm, which is 80% of 2300lm achieved by the SFT42R; I consider that fairly close. The 25R has 2.25mm^2 of LES compared to the 42R’s 5mm^2, and would thus achieve greater intensity by more than a factor of
1850/2300 * 5/2.25 > 1.78,
which is a substantial improvement in my book. Furthermore, there were several reports of the new 25R batch noticeably outperforming the one tested by koef3, for reasons unclear (perhaps lower Vf, perhaps higher efficacy). In that case the lumens gap narrows, while the throw gap widens.
Under identical conditions (fan cooled to 25C), the XML2 actually outperforms the SFT42R in output at 5A (where the test stopped), achieving 2050lm compared to 42R’s 1850.
The advantage of modern LEDs is mostly in power density and overdrive potential (which are really equivalent notions measuring how much power you can push through a given area). At moderate drive currents, this advantage narrows and first-order effects (like extraction efficiency difference between domed and domeless) become more apparent.
Modern LEDs are also more efficient in a subtle way–at the same current and similar output, the Vf is often much lower. However, this efficiency advantage is not detectable in a lm vs current graph, and often isn’t relevant to peak output, since drivers are current-regulated rather than power-regulated. (FET drivers may destroy a low Vf LED on a high-drain cell.)
I’m very curious why this is the case, as I recall other tests of the old XML2 also failing around 5A. Perhaps there were already different batches of the emitter floating around, and the K40vn happens to have a good (or hand-picked) sample?
Chances are that this (and the above failure at 5.3A) is mostly attributable to the few/thin bond wires rather than a fault of the actual die. The mode of failure in koef3’s test clearly shows destruction of the dome around bond wires.
