Vega's Integrated Graphics performance

Yes, your summation is mostly true except for the scale of magnitude between integrated gpus and dedicated gpus of even modest middle range.

The number of compute units or shaders that are being used for compute is an order of magnitude of 10X-100X in favor of the dedicated gpus.

And number of compute units performing the calculations is more important than power usage with regard to crunching times. Simply the scale of the parallelization if the application is written to correctly use all the available shaders.

A integrated gpu might get through a task in 1200 seconds for a modern igpu but a dedicated gpu would process the task in 120 seconds.

You just have to decide which is more important to you, crunching speed/RAC or efficient power usage.

The problem with integrated gpus is you have to share the memory and compute resources with the cpu and system.

I see too many posts of igpu users complaining of very long crunch times simply because they load the cpu up with 100% utilization on cpu tasks that starves the igpu of memory and compute time. If they simply reduce the cpu load down to 50-80% or similar, the igpu crunch times improve greatly.

You do save on the cost of the host computer though with not having to purchase a dedicated gpu. But the number of projects that offer igpu applications is not great.  Projects tend to have more apps for dedicated gpus than igpus so that limits your project selection choices.