That’s an interesting question. I would expect that combining the two GPUs in a single run would not benefit you, even though their clock frequencies are quite similar. Running two or more simulations in parallel should work, however, separating the workload on the GPUs. In this scenario you should however take care to pin the right CPUs to the corresponding GPU, hwloc should help you here. Maybe @pszilard or @zhmurov have some more thoughts on this.
I agree with Christian: It will be nearly impossible to balance the load between two GPUs that are two generations apart. However, you can always run one trajectory per GPU, using 1080ti for system preparation or for smaller systems. Check-out -update gpu feature in GROMACS 2020 that can help you to use two GPUs in one system more efficiently.
You can still try using the 1080Ti for PME work (and you’d likely have to use the GPU direct communication feature, but you may still not see a large performance benefit (especially if you have a consumer motherboard which will work only in PCIe x8 with 2 GPUs).