The reason I have 5 is because there’s lots of transient spikes on the NVidia cards and I can’t run 3 cards on a single 1600W psu without stepping their power levels down and I didn’t want to do that. While most PSUs have an 8-10% buffer of overage, I tripped my 1600W PSUs constantly on the transient spikes even with stepped down cards at 500W x3, fully loaded they will go above 600w per card sometimes, and also 600x3 = 1800w, which the math doesn’t math so I put 2 per PSU.
My setup is 2 GPUs per 1600w PSU + the MCIO boards which have a 35w spec. So roughly 1270w per psu under load, in reality they sit closer to 1300w a piece when fully loaded. I was going to run 4 1200w for the cards but we’re in a supply chain lull right now and I simply couldn’t find the ones I wanted because they aren’t making them anymore and I already had the 2 1600w units running and I like stuff to all be the same.
I have a 1200w PSU for the motherboard / cpu and PCIe bus because the retimer cards draw some juice too, plus fans, water pump etc. So that’s the “main” PSU and then the 4 PSUs for the GPUs, at 2 GPUs per PSU + MCIO boards. It works, it’s stable and I can run full load with transient spikes all day long and it doesn’t trip anymore.
