I recently installed Centos 7 on a Dell PowerEdge R740 with a Nvidia GPU added to it.
After installation I was surprised to see that the iDRAC remote console capabilities was not working.
Not even console com2 when doing ssh to the iDRAC even though there is a serial-getty@ttyS1.service running.
Getting the serial interface working was just a matter of switching the port mapping in the iDRAC from serialportaddrsel=serial1com2serial2com1 to serialportaddrsel=serial1com1serial2com2 using the iDRAC gui or the command line tools (syscfg/racadm).
But fixing the remote console in the GUI is another story.
After some dialogue with Dell I was able to confirm that this is the expected behavior as the remote console only shows what's on the integrated GPU and the OS only sends output to the Nvidia GPU.
The subject of ttys is a bit mystical with it being layer upon layer of abstractions emulating long dead hardware, but should it not be possible to have ttys presented on all GPUs in a system?
Doing lspci | grep VGA lists both GPUs, so my interpretation is that the hardware allows linux to use both.
Best Answer
Normally, when only one GPU is being used by the virtual console TTYs, there is just one virtual console TTY actually usable at any time, and so there is no ambiguity as to where any keyboard and mouse input must go.
But if you have two active GPUs and virtual consoles on each, you would need to have some means of switching your keyboard (and mouse) focus between the two active TTYs. The virtual console TTY subsystem does not do this (as far as I know...), but a GUI desktop can cope with it, as switching mouse and keyboard focus between things is a perfectly normal and expected operation within it.
Or... you could plug in a second mouse and keyboard, and have two completely separate Keyboard-Video-Mouse sets. This would be a multiseat configuration.