I have some performance information for the 18 core iMac Pro based on a integer/branchy workload (CPU limited compiling with all cores in Xcode).
I derived this information by building with 36 processes in Xcode, and then disabling cores in Instruments. This probably means that it's not entirely representative of a real x thread workload, because only some parts of the CPU is taking all the thermal load. A real workload should perform slightly better due to more even thermal load.
The max turbo boost for X number of cores enabled:
View attachment 750798
I started the load when the CPU was at 50 °C or lower. It seems that the maximum sustained power allowed is 133 W, so anything below that will maintain maximum turbo boost indefinitely. For 6 core or less the maximum turbo boost is maintained indefinitely.
For higher core counts the CPU will briefly use more than 135 W, all they way up to 170 W. For my workload this brief period was maintained between 7 and 74 seconds depending on the power usage.
View attachment 750799
Aggregated frequency, simply multiply frequency by number of cores:
View attachment 750801
After the brief period of maximum frequency, for higher core counts the CPU will begin throttling to maintain 135 W power usage:
View attachment 750803
Over time the frequency will will decrease slightly for higher core counts as the temperature increases until eventually the temperature reaches about 88 °C and a steady state emerges:
View attachment 750804
View attachment 750805
Aggregated frequency, simply multiply frequency by number of cores:
View attachment 750806
Some conclusions comparing to
https://forums.macrumors.com/threads/imac-pro-18-core-delivery-status.2100673/page-2#post-25715226:
(Assuming max turbo boost for 10 core with all 10 cores enabled is 4.1 GHz)
When you have a highly threaded bursty workload. For example editing filter settings in an image editor, for the best case in favour of the 18 core the 10 core could take about 40% longer time to produce a result (18 has a 29 % shorter response time):
(3.2 * 18) / (4.1 * 10) = 1.40
In reality the difference might be less.
For a sustained workload such as a long code compile you could expect the 10 core to take about 30% longer than the 18 core to finish the compile (18 core take 24% less time to finish):
(2.6 * 18) / ((3.49 + 3.64) / 2 * 10) = 1.31
So in conclusion
The 18 core can lower your response times up to 29% vs the 10 core
The 18 core can lower your wait times up to 24% vs the 10 core
Conversely, some single threaded and lightly threaded workloads could take 4% less time to complete on the 10 core, but this is somewhat reduced by the larger L3 cache on the 18 core.