There’s been quite a bit of hype over AMD’s latest processors. The consensus seems to maintain that gaming should prioritize higher core speed over higher core count, studio work vice versa. But, the latest processors appear to be closing the gap.
This got me wondering: just how well does DOTS utilize / scale with high core count CPUs?
If DOTS pairs quite well with high core count CPUs maybe Unity and game devs have a nice marketing opportunity here. Would love to hear your thoughts.
It scales really well. You still need enough single-threaded performance for the main thread to finish its main-thread exclusive work, but that is also becoming less of an issue.
Unity is marketing this in the server space where it saves developers server costs. It is up to developers to market their games to high core-count users if they want to segment their audience that way.
I’d buy one of those 32 core beasts but I’m afraid I’d accidently make my game run like crap on a 4 core.
That’s a good point. Maybe we could turn off all but 4-8 cores and reduce the frequency to get some idea for older CPUs, but I don’t know how well that translates.
I think 4.5-4.7 Ghz was the overclock on some of the new AMD processors, which is far more than what I got with an i5. Thus far the DOTS demos have been running extremely well on this old machine.
If you do it in the UEFI, it translates almost perfectly. I think Unity was also looking at setting a threadpool size in software. Not sure how accurate that will be with background loading threads.
That depends on how much main thread work you do and what your target framerate is. For some projects and teams, that constraint will be trivial to make a non-issue. Others might struggle. And I wasn’t saying that Ryzen 3rd gen was weak in single-threaded performance for the main thread. It is definitely not. If your game runs better on latest Intel compared to latest Ryzen at under 240 FPS, you are probably doing something else.
I am not sure that gimping a few cores on the test machine is any kind of reliable indicator. Best to just grab as many old bits and pieces lying around and build some boxes that cover the low, medium and high recommended machines in the steam hardware survey. Failing even that, could get old laptops, or beta testers and friends.
I know the friends part is hardest for us developers but we must try 
“Hey mom, yeah I’ll be in town soon. Say, can I borrow your laptop real quick?”
The tech media industry over the last few years has been taking flagship processors and tweaking the UEFI to match the core count, SMT settings, and clock speed of yet unreleased lower-end products of the same generation. Their benchmarks are almost always within 1% of the actual results when the product launches. And this makes sense too as the lower-end products are just higher-end products with things disabled because the fab wasn’t perfect. Intel uses fuses to disable stuff in hardware. AMD use to disable it in software (hence core unlocking was a thing) but are now rocking their chiplet design.
That’s not to say that you shouldn’t profile on different generations of hardware and on mobile versus desktop processors.