We just purchased a 4 M2090 GPU server with Supermicro X8DTG-QF+ motherboard. The motherboard specification show that it has 4 PCIE4.0 X16 slots. But after I did “nvidia-smi -q” I have varying results from the 4 M2090 GPU’s
The M2090 with ID:0 shows:
Max : 2 Current : 2 Link Width Max : 16x Current : 16x Power Management : Supported Power Draw : 73.06 W Power Limit : 225 W
while the M2090 with ID 1, 2 and 3 show:
Max : 2 Current : 1 Link Width Max : 16x Current : 16x
Power Management : Supported
Power Draw : 28.32 W Power Limit : 225 W
There is a difference between the PCI slot generation and the POWER DRAW at idle. As this is a new setup we want to be sure if everything is correct or not.
Please can anyone shed some light on what the issue maybe ? Can it be a power supply issue ? or is it a motherboard or a software issue ?
Also how much of the performance difference will this cause in application where we have lot of CPU-GPU mem trasnfer ?
The configuration of rest of the system is as follows:
2 Intel hex core X5650
48 GB RAM
RAID 5 ( 6 drives ~ 500gb each)
Scratch drive - Intel 120 GB SSD