For first rig(since your not running VMware), I would run 2 instances of SMP and the GPU client. You will need to set affinity up, as well as priority for the SMP and GPU clients sharing the same cores.
Second rig. One instance of SMP as well as the GPU client. You will need the priority here as well.
Second rig. One instance of SMP as well as the GPU client. You will need the priority here as well.