S4MAX: Brickless S4M w/ 2080ti and R9 3900x - 600w - on water

Choidebu

SFF Guru
Aug 16, 2017
1,035
995
Just happen to have some free time after lunch...

Am I correct that this spec's ps_on is just like atx?

So in essence you can just find a picopsu that exposes atx 24 pin's ps_on and connect those to corresponding pin on the supermicro and it'd work?

Have you seen @Thehack 's 24-pin-to-7-pin, M2427 project?
 
Last edited:
  • Like
Reactions: petricor

petricor

SFF Lingo Aficionado
Original poster
May 12, 2018
89
357
Edit: disregard below. I counted wrong number of pins.

I'm 90% sure it follows an OCS Power Supply Spec V2.

Spec page here on OpenCompute wiki

Download page here..
OCS Open CloudServer Power Supply V20.

For easy reference..

Oh you've got updates from another forum, that's great. Yep when I skimmed the pdf spec, that ps_kill jumps right to me. For some reason I thought it won't be as straightforward.

Well the spec is there, will come back to it this evening when I got the time.
Would have been too nice though - thanks anyway! Waiting for a bunch of resistors to arrive to run some tests on the pins...
 

petricor

SFF Lingo Aficionado
Original poster
May 12, 2018
89
357
Just happen to have some free time after lunch...

Am I correct that this spec's ps_on is just like atx?

So in essence you can just find a picopsu that exposes atx 24 pin's ps_on and connect those to corresponding pin on the supermicro and it'd work?

Have you seen @Thehack 's 24-pin-to-7-pin, M2427 project?
Actually it’s what i’m hoping for- my rough and completely unverified plan is to run the +5v standby power from the PSU directly to the motherboard (that implies detaching the 5v output from the pico) assuming that this allows the PS_ON on the MB to be triggered by the power button and re-routed to the PSU along with PS_KILL. If signals are not compatible (again, ZERO documentation on the Supermicro) it might have to involve some switching logic.
I haven’t been aware of @Thehack ’s solution and it seems to do exactly what I need... thanks for pointing this out!
 

Choidebu

SFF Guru
Aug 16, 2017
1,035
995
Ps_kill just needs to be pulled to ground with a 100 ohm resistor so you can do that on the psu. There is no ps_kill on atx standard. It's used for hot plugging.

Unfortunately afaik most server psu only provides 12V and 5VSB (sometimes they even use 3V3 for this), so to conform to atx you'll need -12V (not sure how prevalent this one on modern boards), 5V and 3v3 rails. I think M2427 just routes this from the psu.
 
  • Like
Reactions: petricor

Windfall

Shrink Way Wielder
Nov 14, 2017
1,837
1,358
Nice squeeze with that PSU, but I am curious if you're going to have room for the mobo in there....
 
  • Like
Reactions: petricor

petricor

SFF Lingo Aficionado
Original poster
May 12, 2018
89
357
Nice squeeze with that PSU, but I am curious if you're going to have room for the mobo in there....
Ps_kill just needs to be pulled to ground with a 100 ohm resistor so you can do that on the psu. There is no ps_kill on atx standard. It's used for hot plugging.

Unfortunately afaik most server psu only provides 12V and 5VSB (sometimes they even use 3V3 for this), so to conform to atx you'll need -12V (not sure how prevalent this one on modern boards), 5V and 3v3 rails. I think M2427 just routes this from the psu.
Hypothesis is to have a pico 160 in the middle and only bypass 5v and PS_ON- the pico should take care of the other voltages required.
Will get a PSU tester to run a sanity check on this hypothesis though as this might turn into a rather expensive experiment with a live board!
 

Thehack

Shrink Way Wielder
Creator
Bronze Supporter
Mar 6, 2016
2,245
2,688
J-hackcompany.com
M2427 may work.

1. 5vsb is on a separate line.

2. Ps_on is on a separate line. Uses atx logic.

3. Power_ok is on a separate line.

You can splice the stock wiring to wire it as needed. Stock wiring is 180mm long.

Those 3 inputs are designed to use atx signals.

The unit generates a separate 5v and 3.3v rail for the 24 pin when the PSU is switched to on.

-12v is legacy and I believe only motherboards with Rs232 (old school serial) needs it. It is noted as optional in Intel psu guide, and I can't find what it is used for besides that. M2427 does have the -12v but I am testing future revisions without it.
 
Last edited:

Choidebu

SFF Guru
Aug 16, 2017
1,035
995
The unit generates a separate 5v and 3.3v rail for the 24 pin when the PSU is switched to on.
That's what we wanna hear ^^. Now we go spec hunting! Don't fail me today, my google-fu!

Edit: oh btw, this is just postulation from reading the atx spec (the old 20 pin one not 24 tho), 3v3 vsb should work too. Although you might lose usb standby if it's just routed through. Or WakeOnLan etc. Who knows.
 
  • Like
Reactions: petricor

Choidebu

SFF Guru
Aug 16, 2017
1,035
995
New info I found: it's a rebranded Compuware CPR-6011-2M1, no luck on finding specific datasheet so far. It's supposed to conform to Intel Common Redundant Power Supply (CRPS) form factor, but I can't find spec for it either.

The sku is dead, in the same CRPS line they carry the lowest wattage is 750W..
 
  • Like
Reactions: petricor

Curiosity

Case addict
Silver Supporter
Apr 30, 2016
475
529
Not sure if this will actually be helpful, but I came across this in my search.

 
  • Like
Reactions: petricor

Choidebu

SFF Guru
Aug 16, 2017
1,035
995
SHOULD ANYONE HAVE ANYTHING RELEVANT ON RECORD - PLEASE BE IN TOUCH!

Plugging it in and assuming that the bigger pins carry major loads, I could quicky detect the +5V standby rails and ground - that's unfortunately as far as I got for the time being.
My hunch is that the pins with notches on them carry special meaning - if we count them from the first pin to the last one it's 7 pins - just like many specs I've seen. Then 4 pin after that, could be an extension to the standard - other voltage rails, just like 5V that you found.

Here's some pics I can gather...



So referring to your pic, first pin with the notch could be PS_KILL, pin B25 on my first pic. This needs to be grounded with 100 Ohm resistor.

Then on the other side, pin A25, should be PWR_OK, can be tested with an led and 330 Ohm resistor to gnd, see if it lights up when plugged.

Now A24 and A23, remote sense pins, sometimes can be left as-is, but if it doesn't work, try connecting them with a 100K resistor. try connecting them with +12V and gnd respectively.

Then lastly PS_ON on A21 to test if it all work out.

Best of luck!
 
Last edited:
  • Like
Reactions: petricor

Choidebu

SFF Guru
Aug 16, 2017
1,035
995
Another one to try...
Hopefully hotlinking is fine...



From here

Maybe first establish which big pin and which face is GNDs..
 
  • Like
Reactions: Curiosity

petricor

SFF Lingo Aficionado
Original poster
May 12, 2018
89
357
A few online orders later I'm looking at this:

- A brand new Noctua NF-A4x20 PWM
- And a whole bunch of resistors for pin detection using the elimination method here (great guide and can't link that one often enough... and many thanks for all the suggestions re pinouts received in the meanwhile: The result is indeed very close to the one found by @Choidebu above!)

...time for some rather entertaining experiments!

Using 510 Ohm resistors for pin detection...


...I connect the six possible pins remaining an option after having eliminated all others as per the method above and connect them to ground.
This does look like defusing a bomb and is probably equally exciting, as until now I don't know whether I have already turned my PSU into a rather pricy piece of junk! With all possible signal pins pulled low, the unit should now fire up when connecting it to power...



...and to my great relief it does so - producing a remarkably spot-on 12V output on the load pins!
Next step is removing connections one-by-one to find PS_ON and PS_KILL.


...aaaaand the winner is: Yellow!
Now, that is a bit of a surprise, indeed, as this unit is from a redundant set and I expected it requiring both a PS_ON and PS_KILL pin being pulled low to start up. Apparently it must employ a switchover logic not relying on a PS_KILL - a single pin grounded is enough to start it up, making things a bit simpler.


Next thing I try is to see whether is starts without a resistor - according to the various specs I found and which have been shared here, the resistors shouldn't really be required but during the discovery process (to avoid damage to "wrong" pins), and indeed: I can just directly pull it low by connecting it straight to ground.

The control signal comes at a "high" current of +4.66 V...


...and at 0.03mA.


With this out of the way, I can get to tidying up my improvised interface (haven't settled on the final connector yet - will probably go for a 6 position screw terminal with 1x signal, 1x +5V and 2 GND and +12V each).


Next point of order is fitting the noctua fan:


It should nicely fit into the niche remaining in the case after cutting it to lenghth...


...but requires some work to sit flush and squeeze out every space saving possible:


In order to achieve this, I'll have to mill some pockets into the frame allowing for the holding brackets to come in a bit:

This looks promising - it's requiring rathe precise cuts not to detach the arms holding the motor in place.

And: fits!



Now I need a hole to fit the status led (it was placed on the end I have removed from the PSU previously) - this position here is where I see enough space on the inside to route it to.


A 4.5mm hole does the trick and allows me to fit the original LED mounting plug:


Final step is trimming the status LED lead and fan wire to length (there is no space left within to hold excess wiring):


This looks ready-to-go:


And here's the result: A pretty compact and robust unit...



...that will be fit with a 3pin DC connector in the 7mm gap between fan and case. The TE micro connector I had in mind originally doesnt fit (the smallest one with 250V rating is 8.5mm), requires a bit more research...

Time to connect and give it a final test


Spins up and is basically non-audible! Bit of a worry is though that the original fan is rated at .75A, whereas the Noctua comes at .05A - if that linearily translates to airflow, it would certainly explain the great acoustics but may get me into cooling trouble...

...that said, for now i'm rather pleased with the result giving me the 600W and enough room for the radiator I need to get this to work.
 
Last edited:

Choidebu

SFF Guru
Aug 16, 2017
1,035
995
Now we just need to rush @Thehack on his M2427! It is September after all...

What plan do you have on pcie riser? Hardly any room beneath the tubing, let alone a 90° bend...

A bit unrelated, after all this time I just noticed that S4M's side panels doesn't have any screws on it, only the sides. That does help on fitting stuff.
 
  • Like
Reactions: petricor

petricor

SFF Lingo Aficionado
Original poster
May 12, 2018
89
357
Now we just need to rush @Thehack on his M2427! It is September after all...

What plan do you have on pcie riser? Hardly any room beneath the tubing, let alone a 90° bend...

A bit unrelated, after all this time I just noticed that S4M's side panels doesn't have any screws on it, only the sides. That does help on fitting stuff.
I’m looking at @guryhwa ’s gflex cable used in his Mr. Haru build. Basically as flat as the Netherlands a pancake and the most promising I could find thus far. He confirmed it’s 250mm long- will have to do a little unrolling exercise to determine the precise length required as I have little space for any excess.
The 90deg rotation will require a sharp 45deg fold in the cable which again may require some tinkering to not damage things beyond usability...
 
Last edited: