Well, I think the 7900X had no Mesh OC applied and was at 2400mhz, with the memory at stock, this is likely what’s causing the bottleneck.
Turning off hyperthreading, upping the clock speed, overclocking the mesh and using XMP on 3200 CL14 RAM would have been what I would have done first to see if any performance can be squeezed out. If that doesn’t help, that’s when a platform upgrade is sorely needed.
My 10920X is currently running 4.7Ghz on one core, 4.6 on 5 cores, 4.4 on 4 cores and 4.3 on 2 hot cores. (Nothing throttles when more active cores are used, the cores remain at those clock speeds all the time)
I tuned memory to DDR4-3466 and raised my tREFI to max. (tREFI helps so much with memory performance, but do be careful to only use it at high memory frequencies as if you use high tREFI at a lower frequency you will cause memory corruption.)
I tested my system with mprime and stressapptest. It’s now surviving every cold boot and not weirdly training my memory, and I set a fixed Vcore of 1.24 for the best core vs the cooling solution which is a EVGA CLC 280.
I kinda disagreed with the MSI cooler choice because it’s an inferior pump in the rad design, personally would have chosen the new Phanteks Glacier 240 because it’s the NZXT X53 WITHOUT CAM. (same Gen7 pump, no USB header for RGB, it’s literally a magnetic pump cover with standard DRGB headers that is completely optional)
Most of the way I made my 1080 Ti work was to ditch hyperthreading, up the memory and mesh clock, and do tertiary timing tweaks. I can fully understand though that if it was a workstation machine not built out with the fastest memory that it would be the limiting factor, and having an air cooler with TDP limits maxed out likely still isn’t enough.
Anyways, that’s my perspective after recently upgrading to X299 from X79. I will be upgrading to a 2080 Ti Hybrid in 2022 as I don’t see any sense in getting a 3070 Ti 16GB if they’re going for both sides of the PCB again. (I know not to trust 3090s with poor memory cooling)