Dell AA23300 Server PSU Conversion : 14V.1V/45A Fixed Output Bench Supply

[mnementh]

      



The title pretty much covers it; I got a smokin' deal on a half-dozen of these PSUs made for Dell 1850 servers, so I decided to go in ba**s-deep and designed some 3DP endcaps to make them into a proper (if somewhat loud) bench supply. Maybe more a "chuck it under the bench to get away from the noise" power supply... I dunno.

I've used these and various other server PSUs (did dozens of DPS-1200FBs back in the day before the crypto-weenies discovered them and drove the price sky-high) as high-current DC power sources for a variety of purposes; they're great for powering LiPo chargers, charging batteries and powering old-school lighting, as well as lots of other 12V-powered devices like car audio and mobile radio comms. If you know what you're doing, you can "float" the DC side GND with some measure of safety and stack multiples in series as bulk power for those CV/CC voltage regulator modules and bigger LiPo chargers. That is a "I understand the risks, etc..." mod that we won't be discussing here. Full stop.

This project is only about the relatively safe practice of tricking these things into thinking they're still shoved up a Dell 1850's arse so they'll turn on and tricking them a little more to put out a couple extra volts above stock. The 3DP endcaps I designed here are both around Rev5, so pretty well refined; they do quiet the beast a lot, but you can only do so much with a brick that has a little jet engine inside.

So without further ado, lets get started on the actual build...




First, I have to make a screw hole in the nosepiece to hold it on. I toyed with the idea of just making it a friction-fit, but I decided a screw was more finished-looking. This terror object is a little jig I made of a scrap of hardened wire screwed to Unit number 1; it will locate the screw hole exactly over and over on multiple parts. Yeah, I could have redesigned the part; I decided on the screw after I had 4 of these printed up. Might as well finish them all "the hard way" as waste valuable production time on multiple iterations to get the hole located exactly right. 




I apply a little heat from my soldering iron and the hole is made.




A little more heat on one of the screws I plan to use makes a recessed hole that reinforces itself (these walls are 40% infill) in the process.




Stepped screw and matching hole.




Now to take it apart...




...and the first thing needs to be done is to unplug this cable from the rear-mounted status LEDs. That will be relocated to the front.




Here I'm pulling the LED cable from under the long HV heatsink and routing across to the other side of the chassis.




Next, this fan guard needs to be removed.




These two screws will be reinstalled. This change makes room for the tailpiece to butt up tightly against the back wall of the unit.




Next, I remove the mainboard. There are 4 screws; one hides under the AC plug at the back.




There are 2 screws that hold the pull handle on; they're tucked in pretty tight, but you can get to them without taking the fan out if you have these long skinny drivers.




At this point I take a moment to wipe the insides of the insulator sheets clean with alcohol and to attack fan and PCB with a natural-bristle brush to dislodge most of the larger dusties.




I install one end of the VSENSE Mod wire now as it's a lot easier with that insulator sheet out of the way. I'm using 26ga Teflon-coated wire left over from a Thermistor install on my 3DP. Waste not, want not. 




This insulator sheet always comes loose when you disassemble. Cleaning with alcohol and then (after it's well and properly dried, duh) playing a flame very briefly over the adhesive will reactivate it for a bit...




...so I can reapply it to the side of the fan where it belongs. Yeeg; looks like I missed another spot while dusting with my paintbrush. Bad TinkerDwagon!  Now to reinstall the 4 mainboard screws.




Now it's time to do the final routing of the LED cable around the other side wall and down into the recess for that one screw. This will allow me to...




...reinstall the cover and screws. This last one stays out;  that screw hole will be used for the nosepiece.


*CONTINUED IN FOLLOWING POST*

[mnementh]

*CONTINUED FROM PRECEDING POST*



Here I'm routing the VSENSE Mod wire up to pin rows 3 & 4 so I can cut it to length. I add ~30mm to give me extra for handling the wire and in case I need to cut it back and re-tin.




It's easier to do  the PSU ENable jumper first. First, I'm going to bridge pin 1A with this GND pad right next to it; I'll solder the end of the jumper in at the same time.




The other end connects to pin 6B (Pins 1A & 6B pulled LO). That's it; this PSU will automatically come on and produce power at full capacity as soon as AC power is applied. These PSUs don't completely turn off unless given a kill command by the server over the I₂C bus; without that, the AC side and fans run anytime power is applied, so no point in screwing around with a switch in this circuit. You CAN do this with 0.100" pin jumpers from the front pin matrix as well.



Next is the VSENSE Mod. First, a little Kapton tape for safety's sake.




Next, I solder these two resistors as seen here to Pin 3A (GND) and Pin 4A (+3.3V STBY). These resistors were chosen by trial and error based on a mod loosely outlined in one of my RCGroups threads. The Kapton tape provides a little extra margin for heat expansion as well as insulation.




And the VSENSE wire soldered across the two resistors finishes the mod.

This mod uses a resistor bridge to artificially elevate the (-VSENSE) pin on the reference IC such that it thinks the (+ OUTPUT) is about 2V lower than it really is. A net of +14.1V is the practical limit for this mod; OVP will shut output down if that goes above 14.3-14.5V no matter what this part of the VSENSE circuit thinks.




Now time for the new status indicator LED. First, these two wires in the plug need to be swapped around to match the pinout of the new LED.




I used a RGB LED in Common-anode configuration to match original circuit and because and I had enough on hand for all 6 units. Note that the common-anode (LED + voltage) wire in this case is the black one. I omit the blue element here because it turns the output amber or purple; pretty useless.




Now time for functional testing. Under normal operation, the LED will blink red for a fraction of a second during POST, then turn green when POST completes and powers up. Lookit dat delicious 14.1V power. OLP testing by shorting the outputs results in red LED and shutdown. 




Now time to install the endcaps. The tailpiece is held in place pretty well by this locking clip; you can see the muffler slits pretty nicely here. This revision produces just a little backpressure on the fans. These are meant to have some low-pressure loading on the inlet side; they are configured in a High Volume/High Velocity compressor stack which responds well to this treatment.




The nosepiece is a little more complicated. First, the Status LED needs to be started in the bezel. Here you can see that I do put some heat-shrink on the plug and that exposed blue element lead.




Then slide the nosepiece into place. The front of the PSU presses against the back of the LED & plug, holding it all in place in the bezel.




And now the last screw through the nosepiece to finish it all up.




I toyed with the idea of making the nosepiece a press-fit; I like this better. Feels more finished.


      

And here it is all finished up. 14V @ 45A of industrial-duty DC power. Yum!




Time to go build me another caveman high-current ballast and do some load testing!

mnem

[mnementh]

Caveman Load-Testing Stuph!

            


Or... How Many Angry Pixies in That Stockpot?!?

The short answer is EXACTLY 36,354 as of 59 min 56 seconds. 

Welp, I just inconvenienced a great many electrons load-testing this PSU at ~500W load for an hour... it brought my wet coil load up to a few degrees from boiling, and it hardly seemed to break a sweat. Fan ran constant speed all the time, it came up quickly to a nice stable ~50°C exhaust temp and pretty much stayed right there for the entire time. Loading varied from 575W/42A at start and dropped to 475W/35A as the water heated up, increasing the coil's resistance... but the output at the PSU sockets remained rock-steady at 14.13-14.14V.

As for noise level, actually it was a pleasant surprise. The wife watched TV in the same room without even noticing, and I watched some Linus videos with it between myself and the monitor and found it completely ignorable.

In a quiet lab, trying to listen to faint radio signals... yeah, it would be annoying. Otherwise it is very livable. The muffler endpieces make a great improvement; perceived noise is easily cut in half. My next lab will be to do some proper comparative dB measurements. 

mnem

[Jay_Diddy_B]

Hi,

I did a little reading about these AA23300, I believe that pin B2 is the fan speed control.
It is expecting a PWM input 0 - 3.3V

Grounding pin B2 will slow the fan
Applying 3.3V will increase the speed fan.

I have not tried this.

I just read about here:

https://www.rcgroups.com/forums/showthread.php?2099662-Anyone-here-with-a-Dell-AA23300-PSU-Q-question

It is probably worth checking out.

Jay_Diddy_B

[mnementh]

Yeah, I'm intimately familiar with their much bigger Server PSU modding thread; I used to live there. It has a fair bit more info on the AA23300, but it's scattered around.

https://www.eevblog.com/forum/testgear/test-equipment-anonymous-(tea)-group-therapy-thread/msg3265138/?topicseen#msg3265138
I already chased down that rabbit hole a couple weeks ago when I first got them. ↑ That mod doesn't work on this variant, and the variant it does work on it makes the PSU go into shutdown eventually; usually within a few minutes. In fact, any mod that actually works to slow the fan to a tolerable noise level puts it into shutdown.

The only way to really manage anything on this PSU is via I₂C, which is how the server communicates with it. That would require a lot better Ardu or STM32 Kung-Fu than I have, and a lot more more time than the project is worth. A dumb chunk of moderately high current DC is what it's good for unless you want to connect it to a computer. No need to complicate things.

mnem
 

[RockenRanger]

Whats the key to adding 2 or more in parallel? Everytime I do only one works and the rest shuts down