I understand that even if Vcache massively benefits old games, they can't exactly market their new CPUs with old games; however, given the 5800X3D's unchallenged dominance in MSFS2020 last time, I'm surprised it's not even on the slide for the 7950X3D.
Also, really do not care for the 7950X3D, show me how the 7800X3D performs. No scheduler/AGESA excuses for bad performance with that one. The leaked reviewer guide is all over the board and doesn't even have data for half the games vs. 7950X
IO die is the same, but the changes do generally net a ~5-10ns reduction in DRAM latency with the same memory kit. The UMC itself is not so responsible for latency penalties on the chiplet design
The used the same IO die but made changes to the IF
The latency changes tend to be because of more cores per die - having 3-4 cores at low latency vs 6-8 means you're far more likely to have less latency between the cores, and then they have higher clock speeds and more cache to boost it even more
Playing around with my 3700x has reinforced my views that once you max out a single CCX, you get a performance penalty in gaming - its got amazing performance in engines that use <4 primary threads, but once you pass that you either get the inter-ccx latency penalty or the SMT threads and their lower performance
It's still better than a 4 core alone because the OS and background tasks can use those cores and its' better to have lowered performance than 100% CPU usage, but i've definitely learned to focus on the most cores in a single CCX (or P cores on the intel side) for longevity with gaming performance
yeah a lot of games I play FFXIV + Starcraft 2 for example have a big performance jump with the x3d cache.
I've seen that as well, i've gone back and replayed everything again (all the campaigns and the coop missions) and noticed that i'm not seeing *any* late game slowdown, unless the other player lags out
It used to run atrociously with certain effects on screen, and it's just... not.
Try lowering both your SCLs to 4 or even 3.
Also you can try tFAW 28 which is usually calculated but multiplying tRRDS X 4.
Try all changes separately not at once.
Testing them now for you
I'm finding it interesting that settings either work, or no POST - theres very few settings that POST and are not stable (RAM at 3333 or higher boots, but has WHEA errors - so that's just IF/SoC, and a newer CPU would fix that)
SCL's working at 3 perfectly fine
Will try tFAW 28 after R23 finishes a 30 min run
tfaw 28 is fine too
one of these changes makes the BIOS lag for keyboard input, but no issues in windows. Odd.
Disabling CSM causes this. Known bug from multiple gigabyte boards, OH WHAT FUN.
TRFC lets me change it now, i think part of the issue was the keyboard lag screwing up some adjustments i was making
Updated benchies OTW
not much room left to tune
that way you also don't have to worry about where to route the button, if you like to keep your case closed
