A reasonably large proportion of workstation-type tasks do parallelise very well. The example I'm most familiar with is professional audio. A typical audio proj

A reasonably large proportion of workstation-type tasks do parallelise very well. The example I'm most familiar with is professional audio. A typical audio project will include hundreds of plugin instances, each with their own thread. Performance for this workload scales more-or-less linearly with core count. As I understand it, many VFX and video editing workloads are also highly parallelisable. >Or is it more "Everything is 99% idle until I need to run that Photoshop filter, and then it does it really fast"? Using all your cores all the time doesn't really matter. A Digital Audio Workstation is very much either/or in terms of CPU use - as soon as you stop playback, your CPU usage drops to idle. If you run out of CPU overhead during playback then everything grinds to a halt, which can be absolutely disastrous when you've only got 3 milliseconds of buffered audio. We want enough FLOPS to cope with our normal workloads, plus a substantial overhead to cope with spikes.