The element that particularly struck me here was ScyldWorkstation – a 4k virtual desktop, which might be another solution for visualisation, GUIs, and other workstation replacements and instant-access use cases.
The two elements here were support for instances in the cloud (see mention of CycleCloud in this context from earlier) and, in a more basic sense, a commercially supported version of Sun Grid Engine. The features seemed to be good in the sales pitch, although, of course, cost is a factor to be balanced against functionality.
Of particular interest here is the vGPU option. This can apparently do two things: allow virtual machines and containers access to GPU resources on a host in an efficient way, or also segment a large GPGPU (e.g. a V100) into smaller units to potentially replicate a workstation with a prosumer GPU in. This could be useful to allow users currently using a workstation (low CPU core count, prosumer GPU) transfer their workload pretty directly. In theory they might be able to use a V100 but might not be able to scale their work to that or make use of all the cores on a high core-count system, meaning wasted resources unless other CPU-bound jobs are shared on the system. So segmenting would allow a 40 core node with a V100 but effectively chopped up into half-a-dozen workstation-sized units. There might still be bottlenecks but it might allow purchase of V100s to support power users whilst also supporting users with a more HTC-type GPU workload.
I debated with Univa about the support for such models and tools in Grid Engine.