Working with a Tesla S1070 connected to a host with two HICs and one GeForce, which is effectively giving us five GPUs to work with. We’re doing some work that could really use being able to address each of the Tesla cards as separate X screens for rendering (the idea being to a frame buffer object or something similar as obviously the S1070 has no video out). When we look at the available screens in nvidia-settings however, only four appear: 1 for the GeForce, and three inside the Tesla. One refuses to show up, no matter how we shuffle the HICs, cables, or how many are connected - we’ve tried with only one HIC as well, and for whatever reason, one port on the Tesla always gives us access to two GPUs while the other only gives us one.
Important to note is that ALL the GPUs DO show up under nvidia-settings and via lspci. It’s just that only three of the Tesla GPUs show up as available screens, and one specific GPU within the Tesla is always missing from the enumerated screens.
Do you guys have any thoughts on this? Is it possible one of them has just gone bad? As I said, it still shows up as an available GPU, just not an available display… is there something inherent in the hardware?