I have read about how Windows 10 1903 and above had made Windows 10 Ryzen topology aware where as prior build like 1809 and below were not and it can affect performance based on switching between CCXs and latency penalty? Now I have my Ryzen 5900X manually overclocked with CCX0 set to 4675MHz and CCX1 set to 4575MHz and perfectly stable at each with VCORE of 1.287. Now because I have it manually overclocked, would the Ryzen scheduler fix in 1903 and above not matter? Because would Windows schedulers even before 1903 know to just use 6 fastest cores by default which are statically set anyways all on 1st CCX and it would not have to find them and hop between CCXs as first 6 cores are 100MHz faster always based on static overclock? As per core overclocking with some cores being faster is not a new thing and was case on Intel CPUs as well. So would Windows just know to use statically set faster cores by default as fastest cores are all on one CCX and it would not need to hop over if using 6 cores or less? I also have SMT disabled BTW. Or does 1903 and above still matter in such a situation? I hate non-LTSC versions of WIN10 but do not want to lose performance and there is no LTSC version of Windows 10 above 1809.
Hopefully it is actually good and not cancelled as Microsoft hates people using it. Anything can change at a moment's notice. And hopefully it does not have the stupid overlay system which always prompts for something you ned a new app to open this MS gaming overlay link in Enterprise 20H2 even with XBOX Gamebar r3egistry tweak and everything Ic ould do turned off, it still does it. That's why LTSC is so good. Not sure though if this is something in newer builds that are issue as Microsoft is trtying to force store on you, or just something of non-LTSC versions. And do you know answer to original question? I am sure scheduler fix would matter if I used PBO or manual overlcoked all cores to same frequency. But the fact that I have one CCX faster and one slower, would WIndows in lightly threaded apps know to use faster cores first which should make irrelevant CCX hopping latency as 1st 6 cores on CCX0 are 100MHz faster? The reason for that is I could not get all CCX's stable at 4675MHz or above. CCX0 I could but CCX1 maxed at 4575MHz at least at safe temps and voltages. Never tried to go higher to unsafe voltages as it was not stable faster even at mild high vcore, and did not want to try at too high of vcore. But maybe if Windows 10 1809 without scheduler works out better this way but using faster cores first without knowing they are on one CCX??? Maybe for the better afterall???
Geekbench barely showed a difference on a 2700X with 1809 LTSC and 21H1, but the numbers were slightly higher on 21H1. For real-world usage and gaming, I didn't notice any difference, even with VR. I use 21H1 since I have it tamed enough to be usable without too much effort, but I wouldn't have an issue with 1809 LTSC today.
After some rounds in VR and a few other games with LTSC today, I'm at the conclusion that the scheduler improvements don't matter, at least with a 2700X. Games work fine (maybe even a tiny bit better), and AIDA64 matches scores on the cache benchmarks for LTSC, W10 21H1, and W11 22449.1000. I've forgotten how clean and light LTSC was; feels great to be back on it
Yeah I just stick with LTSC. I have read that Zen 2 and above separate CCXs are not treated like NUMA nodes but use UMA for memory access. Where as Zen+ and below the separate CCXs are to be treated like NUMA nodes. So maybe the scheduler improvements are less important on Zen 2 and above than Zen+ and below? I believe the Core 2 Quads had separate packages with 2 dual cores rather than native Quad Core and Windows 7 scheduler new how to deal with it just fine. I am not sure why Zen would be any different.