Hi here is a benchmark in WoW (Still CPU driven game) comparing the two memory profiles. The settings in WoW where the optimized performance suing 4K on a crowded main city (Donogal). I was expecting 8000 Mhz to have the edge, since PyPrime, Y Cruncher indicated that. What could it be the main driven for the difference?
9 Comments
Timings>Frequency
When you have the X3D cache it often ends up being the timing that is your bottleneck. There are ofc. benefits of having faster MHz as well though.
Different workloads will favour bandwidth or latency.
Your best gaming performance is likely to come from DDR5 6200 + 3100MHz UCLK + 2066MHz FCLK.
1.6 MEM VDD + 1.45 MEM VDDQ should be enough to run CL28 + 135ns tRFC at DDR5 6200.
It should also be possible to tighten your tRRD, tFAW, and SCL timings.
https://youtu.be/iux-P7qGe-o?t=338
I could be wrong on this, but PyPrime is latency sensitive iirc, which means it should benefit more from the tighter tRFC & UCLK:FCLK on the DDR5-8000 setup. World of Warcraft should be memory bandwidth bound, so it likely benefits from the higher UCLK & FCLK with the DDR5-6000 setup.
It will get better as newer cpus comeout as right now am5 memory controller cant do anything above 6200 properly without 2:1
SO, people say its cuz timing is more important than frequency for AM5, which is truem, but because of another reason.
MCLK (memory(controller) clock) and FCLK (infinity fabric clock) are linked, and ryzen 9000 can only do 3100Mhz FCLK usually, sometimes 3133 and 3200MHz, and very rarely higher. So if you use Memory with higher than 31000MHz (6200MT/s DDR) you have to run MCLK and FCLK in 1:2, so FCLK with 8000MT/s is only running at 2000MHz instead of 3000MHz on the 6000MT/s config. Meaning the individual cores are connected with a slower Link, and the additional Memory speed cannot compensate that difference.
In more (directly) mathematical applications, predictions become easier and memory bandwidth becomes more important than how fast the individual cores talking to each other, as they all usually do their own (part of) calculations and are not as dependant on each other
Congrats on lowering performance lol
The UCLK should be above 3000Mhz to get best results
Outside of wow potentially being bandwidth limited (try same fclk on the 8k setup), did you stress test this properly? Might be unstable/performance regression.