-2

I have a server with two 1600W power supply units (a Gigabyte GS-R22PHL if that helps).

Should I assume the maximum rated power is 1600W, and that the second unit is for redundancy, or that the combined 3200W are available to the system? I apologize for the somewhat trivial question, but the manual does not mention anything either way, so I assume some standard practice is in play here.

0xF2
  • 187
  • 9

2 Answers2

1

Generally systems with redundant power-supplies are designed for exactly that, redundancy. When one power-supply fails, is unplugged or whatever, the other must be sufficient to keep the system running by itself.

That means that the second power-supply is not designed to allow the system to draw more power than a single outlet/power-supply can provide.

Bob
  • 5,805
  • 7
  • 25
  • Thank you. I thought as much, but I guess I was hoping ;-) Realistically, this server is going to use maybe 500W all the time, and only when the coprocessor expansion cards kick in push to 1200-1300W – 0xF2 Feb 23 '22 at 13:47
1

The maximum it CAN draw will be 1.6Kw, in reality you won't be able to do that even if you fit the most power-hungry components, in real terms you're unlikely to use 70% of that even if the server is 'pegged' at 100% utilisation of all components 100% of the time.

Most server manufacturers actually list their maximum and average power draws but I noticed that model doesn't list them like that so I'd go with Theoretical_Max=1.6Kw, Realistic_Max=1.12Kw but if you measure it I bet it's mostly pulling about 700-900w almost all the time.

Chopper3
  • 101,299
  • 9
  • 108
  • 239
  • Thank you. I thought as much, but I guess I was hoping ;-) Realistically, this server is going to use maybe 500W all the time, and only when the coprocessor expansion cards kick in push to 1200-1300W – 0xF2 Feb 23 '22 at 13:46
  • Now I'm intrigued - "coprocessor expansion cards" - what's that? – Chopper3 Feb 23 '22 at 17:43
  • That's what Intel calls the MIC/Larrabee/Knight's landing. Marketing name is Xeon Phi. I am using the term somewht liberally as these systems will have both PHI and GPU accelerators. – 0xF2 Feb 25 '22 at 14:22
  • 1
    Oh I know larrabee well, we tested it as an engineering sample years ago, we never really took them into production and moved to nvidia tensor stuff for ML. – Chopper3 Feb 25 '22 at 15:31
  • 1
    makes sense - but now a Larrabee can be bought for $40, it is interesting hardware to teach parallel programming with MPI on the cheap. – 0xF2 Mar 09 '22 at 20:08