Since the article does refer to these as desktop CPUs, I'm curious what kind of desktop workloads people are running that could benefit from / justify them?
Gamedev is a good candidate. Compiling in parallel, running different processes that transform resources, running a not yet optimized game engine in debug build, etc.
It's fun being able to start a bunch of brute force number crunching jobs that peg cores for hours and still be able to use the computer as if nothing's going on. Browse, code, listen to music, watch video, everything runs smoothly.
Intel likes to market these CPUs to people who 'mega-task'.
Video Editing while
Gaming while
Streaming while
Hosting a server while
Dealing with encrypted data while
Multi-monitor while
Contributing to science while
while
while
etc
AMD hasn't really declared which market they are after yet. It's likely going to be a big part of the launch event.
Mega task - as in have a hundred tabs open, a dozen Excel spreadsheets plus Outlook. Maybe there will be lots of cores since there is no more Moore's law, 32 core computing will be the new core i5 with 64 core computing being the new i7. Some software might run albeit slowly on older computers that only have 16 cores.
In this future the choice of 64 core computing is one of those things like car choice. Obviously we need cars electronically restricted to 155, we would not consider a car that only did 120 or so. There is a theoretical scenario where that vital power is needed but it is not a rational decision based on cost benefit analysis.
Therefore you can expect a change over time to where core total becomes marketing.
Heh, IIRC the "mega-task" thing is very very recent, likely in direct response to AMD upping the core counts with Ryzen. (well, technically FX was 8 core but let's not go there)
I used to dabble in real-time ray tracing. My box is from 2005 though so it was never really great. These chips should easily do moderate complexity scenes at 720p and 60fps, and that's just an 8 core. Figure 1080p 60fps with 16 cores. But then we'll go 4K and turn on a lot more features and slow it down to a crawl again...
Path tracing uses a lot of rays per pixel. I do a single eye-ray and a few to the light sources for the surface intersected, and of course reflections. I do have a scheme for handling large numbers of light sources to some extent. I would say moderate complexity definitely includes a near-full-screen object of arbitrary polygon count. It may also include environments like anything from the quake series - including light sources. Of course I need to buy one of these things and try out the old code to see if we're really there yet. Until then I'm still speculating.
I've got a lot of slow build in my life. Large compiled projects are CPU heavy, easily parallelized workloads that offer a great ROI on speeding up. The cost of a new CPU is very little compared to wasted engineer time and broken concentration.