After a little setback (more below) and little time at hand, I’m back with the next Episode of my rather ambitious project: Proof-of-Concept for the power solution.
With the PSU squeezed to measure, it’s now time to verify my hypothesis on powering the build:
Idea is to use a 160w pico PSU to drive the 24pin ATX connector of the main board, but to have the heavy loads, including the GPU and CPU power feeds, driven directly from the PSU’s 600W 12v rail to exploit its capacity without the Pico-plug-in-board becoming a bottle neck.
Also, I’d want to avoid any load switch in the middle - the PSU is platinum rated and has a standby mode, and assuming that the PS_ON is designed to ATX specs, it should be possible to control it in parallel to the pico PSU, using the main board’s PS_ON signal.
This requires, however, to hijack the +5V standby power coming from the pico, and to feed the board directly from the PSU’s standby power feed instead.
As this is somewhat uncharted territory, I have equipped myself with an ATX PSU tester and ATX cable that I can place between pico PSU and board, allowing me to splice in the standby feed and to hijack the PS_ON signal from the board - and to test whether it all adds up to an ATX compliant output, before risking frying a precious board and CPU.
Plugging in the pico for an initial test with the PS_ON connected to ground directly at the PSU, I get a compliant reading- however, just for about 3 seconds, then the PSU switches off the 12v feed!
As it’s impossible to distinguish whether it’s just going to standby or reporting a fault without reading out the PMBus, I’m back to experimentation.
Something clearly doesn’t work as expected.
Either I have indeed missed a second pin on the PSU (PSKill?), which would surprise me as I tested it left, right and centre, or it is somehow related to the Noctua replacement fan I use.
After some quite time consuming testing it indeed turns out to be the Noctua apparently not spinning fast enough, and triggering a fault alert and PSU shutdown.
Searching the web, I have found similar cases reported by users replacing fans on Supermicro boards with slower spinning ones triggering fault alarms.
As there is now way to override this behaviour of the PSU, it appears that for the time being I’m stuck with the original fan, coming with notable noise and most critically: too much length.
Looking at the fan it appears that, similar to the Noctua, I can recess the fittings a bit, and taking a slightly more aggressive approach I aim to gain 4mm which should just about fit - I still have about 6mm left between the psu and radiator to play with.
After cutting back the fittings for the screws at the bottom, the fan looks like this...
...and fits nicely in-front of the standoff for the PCB.
On the top side, where I do not require screws to go in, I simply cut in a 1.5mm slot to interlock with the flanges in the case.
Fits!
Its all a bit tighter now, but it still looks like should be able to just about fit in a 140mm radiator next to it.
To not waste space on projecting screws, I cut sunk holes in to the fan...
...allowing for a flush fit of the screws.
With a new wire soldered to the fan and the PSU re-assembled i get to a rather tight package...
...that finally behaves as desired and runs continuously when switched on! Unfortunaltely this also comes along with a well-audible fan hum - not terribly annoying but miles off what you'd want to have when having been spoiled by noctuas in the past.
And whilst at it: here is the confirmed-to-work pin-out for the Supermicro PWS-606-p i'm using:
...and this image shows the rear side of the connector with the +12V and PS_ON wires soldered on. Forgot to post this earlier...
With fit resolved, there is still the noise issue which I will have to deal with at another point in time- initial research shows a very limited choice of low noise 40mm pwm fans, and it would be trial-and-error as I have no specification for the minimum rpm the Supermicro controller requires.
So, next step is finally verifying my power hypothesis:
Having tested and confirmed the connectivity between the ground input of the pico and the ground output on the atx side in “off” state, it should suffice to route a single wire from the PSU’s +5v Standby output to the matching pin on the boards atx Input to get the board into standby mode.
Using the ATX extension cable, I cut the +5v sb lead (violet), and hook the board side up to the PSU.
Second mod is to connect the PS_ON wire (dark green) between board and pico to the PS_ON Pin of the PSU: If the board successfully powers to standby using the PSUs standby output, it should be able to pull both PS_ON signals on pico and PSU low, 1) powering up the PSU, now feeding 12v to the pico, and 2) directly turning on the pico to feed the board.
I also connect the auxiliary 12v cpu feed Input of the tester directly to the PSU’s 12v rail to test whether it all stacks up.
And- it works! Both PSU and Pico come to life, with a compliant reading on the tester.
That gives me the confidence required for the ultimate test: Using my i7 8700k S4M as a testbed and driving it off my new power unit.
I disconnect both PSUs of the S4M and remove the HDPlex direct plug...
...replacing it with my pico bypass cable, and connecting the auxiliary 12v CPU connector to the supermicro’s 12v output.
With this setup, the board AND the PSU should go to standby upon connecting the PSU to mains...
Success...
...and now power on both PSU and board when pressing the button:
And it spins!
Looks like I now have a reasonably compact, working and a load switch free solution for my build, using the Supermicro PSUs standby capability, and allowing me to feed up to 600 watts directly to CPU and GPU from a single source. Major box ticked.
Next: Modding the pico to do away with the ATX extension and feeding it off four wires, directly from the PSU - and finding a better fan to replace that screamer...
