DIY 8-card platform has card loss

Graphics / Linux Linux
1

I used the AOM-SXMV quad-card v100 16G from Super Micro to build two sets of quad-card systems, totalling 8 cards. I used the X10DRG-Q motherboard from Super Micro, with both sets of AOM-SXMV connected to the motherboard, and it correctly identified all 8 cards. However, when I used gpu-burn, I encountered screen freezing, and after restarting, an error occurred [Unable to determine the device handle for GPU0000:05:00.0: Unknown Error]. After restarting again, one card dropped out. But when I tested these two platforms separately, they both correctly identified four cards each. When placed together, there were only seven cards left. What could be the possible cause of this issue?

thanks…

2

most likely is PCIe issues. what cable are you using to connect the AOM-SXMV board to the motherboard? must be a fairly long cable to reach, and in my experience, the longer cables are VERY spotty and inconsistent in the PCIe signalling, especially if you use a cheap brand cable.