Synthetic benchmarks are scoring particularly high, but most real-world applications are not. The improvements are task specific and it seems that Radeon VII and 5700-series GPUs exhibit the largest gains. Nevertheless, we want to include all user benchmarks whether they be good, bad, neutral, horrible, or exceptional. This will help to identify the operational envelope -- or set of tasks that do and do not benefit.
To that end, if you captured your before/after timings it would be quite helpful to post the data in a spreadsheet table.