Probably an original, old stock, Wight. One of “the good ones” 
The interesting thing is that it’s using an FET driver with an SO8 FET. That didn’t happen for quite some time. The “original” originals all used a 3-channel buck driver. Next there was a wide variety of drivers in the various clones, imitations, or simply new production units (IIRC). Eventually the JB driver showed up and that was using a TO-252 packaged FET (eg the big one). Eventually a modified version of the JB driver finally switched to SO8, probably due to the cost at that point (SO8 had long been “very common” for commercial stuff by then!). By that time (January 2014 we’ll say) I think that the flood of poor SRKs was well underway.
I figure the lumens per watt will keep increasing over the years until we get close to 300, and batteries keep getting stronger, so anything I build now for maximum lumens will become obsolete. However, if I make something nice now, it’ll still be nice later… just not as efficient as it would be with newer parts.
A nice host stays nice a lot longer than the engine inside it, so I’ve been choosing items mostly based on the host.
Plus, I only want up to 50 lumens most of the time anyway. I doubt that’ll change any time soon. I can do most of what I want with 1x7135 and an attiny13a.
Yes, the tailcap and cells are important factors. This discussion kind of assumes everyone has already installed the hottest cells and bypassed all springs and otherwise eliminated as much resistance as possible in the host. Take it as a given that every other possible factor has been maxed out, and the focus is on only the driver.
So, what’s left to improve? The FET is an obvious target for optimization, and there are still a few models which haven’t been tested yet.
We have models? Parade a few of em out over here if you don’t mind…
Wight, Comfy, you might be interested in this…
2 Eagle Eye X6 hosts, all pieces parts to start with. 2 A17DD-S08 boards, ATTiny13A flashed with A6 firmware. (I know, I know) 2 of the XP-L V6 1D out of the same tape, side by side. Same wire length, 20ga Turnigy, everything built identically. Same sized bits of solder paste when putting the switch on the pcb, same goes for the springs. All stock, NO bypasses whatsover. The 2 driver boards had to be separated, as did the 2 SinkPAD 20mm stars, everything was from the same tape, side by side, as identical as I could possibly make it.
2 brand new never put in a light Samsung 30Q, rested at 4.20V.
SIR800DP
5.96A for 1538.7 Lumens
1407.6 at 30 sec
2.71A for 793.5
0.97A for 314.99
0.10A for 42.88
0.01A for 3.07
0.01A for 3.07
SIR404DP
6.01A for 1607.7 lumens
1476.6 at 30 sec
2.74A for 821.1
0.99A for 327.75
0.11A for 44.99
0.00A for 0.828
0.00A for 0.828
My meter wouldn’t show anything on the lowest modes of the SIR404DP, stayed solid on 0.00, which is unusual.
No single 7135, no moon mode, each light has 6 instead of 7 modes, each light does the same on level 2 and 3 , in other words, 1 is blank, 2 and 3 repeat, then 4,5,6,7.
What PWM values for each mode?
From the A6 C file…
// Mode group 1
#define NUM_MODES1 7
// PWM levels for the big circuit (FET or Nx7135)
#define MODESNx1 0,0,0,7,56,137,255
// PWM levels for the small circuit (1x7135)
#define MODES1x1 3,20,110,255,255,255,0
Yes, I know, that doesn’t add up. Makes no sense. According to the PWM levels in the C file, there should be only 4 modes with no 7135 for the small channel. BUT, that IS what I flashed the MCU with, and it IS set up that way, and both lights DO make low level light on 2 and 3. Yes, 1 is blank, as 2 and 3 should be with a 0 setting, but this is what both lights do.
I seem to remember reading in some threads that the fast PWM with a value of 0 is still on for a short time. Some people were using it to get some sort of moonlight mode from the FET. The first mode is set to phase correct PWM so it doesn’t have that problem.
Ah, but of course! The phase correct vs fast pwm settings would explain the difference. Thank you! Thought I was losing it (y’all don’t need to go there. )
Keeping it pertinent, I built a new driver for one of the X6’s, this time I was paying attention and used the A17DD-Hybrid with FET+1 architecture. Again, A6 firmware with ATTiny13A, Samsung 30Q fresh off the charger at 4.21V. 20Ga leads but this time I bypassed springs and used the through board bypass at the switch. 7135 chip is a 350mA version.
0.01A for 0.483 Lumens
0.03A for 9.177
0.14A for 69
0.47A for 223.215
1.36A for 534.405
3.10A for 1055.7
5.94A for 1783.65
30 sec yields 1669.8
This is still the XP-L V6 1D on 20mm SinkPAD, domed, and using the SIR404DP MOSFET. 1783 is pretty potent! 
Nice X6, for sure.
Is the 404 still OK as the voltage drops?
So far, I need to run a full cell killing timeline.
First, I got the new driver boards in. So, do I use the ATTiny25 with these and the Bistro firmware? Any special instruction before I build the first one?
Second, I just ran a kill test on the X6 with SIR404DP. One hour, I quit, Cell was still at 3.10V. All functions normal. Moon was stupid low…
15 min 3.81V 4.75A 1462.8 lumens
24 min 3.66V 3.55A 1255.8 lumens
38 min 3.45V 993.8 lumens
55 min 3.16V 1.2A 538.2 lumens
60 min. 3.10V…numbers as follows
0.0345 lumens
6.6585
54.722
174.225
228.755
320.505
455.4 lumens
With everything working as it should be, moon and all reverse hidden modes working fine, reverse press timing seemingly staying constant, I call it a win for the SIR404DP. One hour and I see why I don’t do this more often! 
I was just worried/curious about those gate specs for the 404 and if it would maybe fall off a cliff as the cell dropped. I think some of the FET candidates have had that issue, even the old 70N02 drops off pretty badly by around 3.5v. But it looks to be OK.
It was still drawing enough to keep the finned X6 head quite warm even up to an hour.
This one has a 45 second step down, so I’d click it back up every time it dropped. At first, about 12 minutes in, it was getting too hot to hold… took it outside where we have a stiff cool wind blowing and ran it til I could hold it. Then just kept clicking it back up til I was sick of it. lol Throughout, I’d run it down to moon, reverse through the modes, then back to Turbo. Stayed working fine all the way.
Right at 1000 lumens 38 minutes in, that’s pretty awesome! This was the Samsung 30Q, button top.
Held it the entire hour, clicked it back up immediately literally every time. Outside, blue sky, big sun… dang light was making a spot on the concrete you could see in full daylight even 30 minutes and more in!
Ok, I’m looking at the component list for this driver and seeing that Alex built one with the ATTiny13A to start with and bypassed the Zener Resistor (200 ohm) . What UI then was used?
And this makes me wonder WHY the 13 components instead of 7? I’ve built a LOT of these with zero issues, so what are we trying to fix if staying with the 13A and normal A6 firmware? If all this is to make it work with the lighted tail cap, leaving that out is far easier!
Even a factory ATTiny25 driver with Bistro runs my Quad Cu X5 with ease, making right at 3000 lumens on the little 14500. I had the tail cap down to a 0.14mA draw with 2 LED’s and a 12K resistor, but removed that this afternoon.
So I’m trying to figure out what firmware to flash, which MCU to use, and just why all the xtra components…. Has R5 been established? Is 1uF still good for the OTC? ATTiny25 and the latest Bistro? Got parts in hand, ready to go, need to know where before I start driving.
In theory, any firmware should work. I don’t think anyone has actually done it yet though, so I wouldn’t expect the best results without a few trial and error steps in the middle.
It doesn’t have to be bistro, but if you go with tiny25 there aren’t a lot of other compatible options yet.
I don’t even use bistro on lights I build for myself. It’s like those big cafeterias you go to when the whole extended family is together and you need dinner but nobody can agree on anything. It has enough variety that everyone can eat something, but none of it is as good as what you’d get at a more specialized restaurant.
What if one used a capacitor between the FET gate and ground? That would smooth out the signal to turn the FET on, which would reduce transient spikes and make the FET operate in linear mode at least some of the time, absorbing some of the energy that is going to the battery and the LED now. The LED and battery would heat up less and operate more efficiently with a more nearly constant current. It would not effect turbo mode.
In the BLF A6, for instance, LED heating in the higher PWM modes is quite significant because of the small head. I am not sure how much it would help the battery but it would certainly be better for it.