Dual CCD Ryzen 5000 CPUs - how bad is latency penalty when threads cross over to other CCD - Is it a big concern for thread heaving games and others

[Wolverine2349]

Like for gaming mostly but some productivity work that benefits from more cores?



Most games do not use many cores and most max out at 6-8 cores so they stay on one CCD. However in the rare games that do scale more, is it really bad if a thread gets caught on the other CCD?? Like a much bigger concern for 5900X than 5950X as there are 2 6 core CCDs and a game is much more likely to scale beyond 6 cores than beyond 8.



So is a single CCD 8 core much better or an Intel 12th gen 8 core with e-waste cores of course disabled in such situation?? Or has that beer fixed that crossing CCDs is minimal latency penalty in games if it happens??

Like could a game crossing over to the other CCX cause a bad FPS drop that causes a pause or lag?? Or is that not a concern anymore. I mostly game but the 5900X with extra cores is great for productivity tasks some of which I also do.

https://www.amd.com/system/files/documents/ryzen-master-quick-reference-guide.pdf

It mentions using Game mode on Ryzen processors offering more than 8 cores.

Someone had mentioned there is a 400% increase in latency if a game thread gets stuck on other CCD?? Is that true or is it minimal difference these days with Ryzen 5000??

News - AMD Zen 4 Ryzen 7000 Specs, Release Date Window, Benchmarks, and More

Page 4 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.

forums.tomshardware.com

 

[Wolverine2349]

It doesn't matter anymore.

The inter core latency was only a problem on ryzen 1000/2000 and the first Threadripper. With Ryzen 3000/5000 and Threadripper 3000/5000 it's no problem and games perform the same whether you disable CCX's or not - the issues were resolved partially by AMD and partially by Windows through better scheduling

Awesome thanks so much. Someone defending the e-cores claimed that 5900X would be worse if a thread got stuck in another CCD as opposed to on an e-core claiming latency between CCDs which kind of scared me a bit. Though as you said that does not matter anymore as that latency has been fixed and all good cores can be effectively used whether it is on 1 CCD or the other. But something getting stuck on an e-core is bad not because of latency penalty crossing core clusters on Intel 12th gen but because e-cores are just limited and much weaker and ruin slower than rest of main P cores screwing Windows scheduler, and especially bad for gaming all around and lack the instructions sets games depend on it seems.