Thanks for this. Yes, the difference in GeekBench is noticeable.Sure, my own benchmark measurements on my i7 2020 iMac:
64GB RAM running at 2667 MHz: 8740 to 9029 multi-score Geekbench 5, memory transfer speed 28500 to 29600 MB/s Novabench.
64GB RAM running at 2133 MHz: 8240 multi-score Geekbench 5, memory transfer speed 24500 to 25660 MB/s Novabench.
Basically 6 to 9 percent drop in CPU multi-score (which I rounded up to 5 to 10 percent in my previous post), and 10 to 17 percent drop in RAM speed score.
These are only my personal tests, but I tried to be as fair as possible, I had no other apps running in the background, did it after rebooting and waiting several minutes. I was trying myself to figure out if dropping to 2133 made a significant impact.
Of course, real world performance is not same as benchmarks, but benchmarks are (almost) the only way we have to quantify comparisons.
Interpret how you wish
btw, OP can run their own Geekbench and Novabench comparisons if doubtful.
I would comment that running RAM with mismatched amounts in each channel (i.e., not full dual channel) results in even worse CPU multi-scores and RAM speed scores
However, in the real world it’s had to say what it will be.
For memory transfer speed to become a bottleneck, I expect continuous loading of big volumes of new data, like video editing, rendering or benchmarks. I don’t see how normal system work and photo editing could be affected.
Thanks for mentioning the speed drop at the beginning in post #6. I guess we both missed it. You didn’t quantify the drop, and the comment wasn’t written like a warning, so we assumed an insignificant drop. Maybe it is.
Up to OP to keep 48GB @2133 or 32 GB @2667.