2bighorn, if I've understood correctly what I've read, RAM speed isn't solely about latency. I've seen several ways to calculate a comparable digit but since I can't remember any of them I googled and this is what I found: https://notkyon.moe/ram-latency.htm. According to that there's a significant difference with 3600/19 and 2133/15, 10.56 vs. 14.06 ns.
I won't argue about the default SPD since my knowledge about it is weak. You're also right in saying that the speed of RAM doesn't matter much.
What I do know for sure is that underclocking RAM allows for lowering the CAS latency so the higher speed RAM may be faster even at 2133.
Bizman, what I was trying to say is that if you buy higher rated memory and just install it into your motherboard, it will run at JEDEC specified SPD profile which is currently at 2133.
If you want higher speed out of your memory, you have to explicitly set it in your UEFI/BIOS, either manually (timings and voltages) or with embedded XMP profiles.
When you do that, integrated memory controller (resides within your CPU package) will run at higher command rate as well (in this case 1800 MHz vs stock 1066 MHz), which means motherboard has to deliver higher voltages to VCC-SA (1.05V default) and VCC-IO (0.95V default) and shouldn't exceed safe voltages for that particular CPU as specified by Intel, which, in case of Coffee Lake is 1.25V (VCC-SA) and 1.20V (VCC-IO) respectively (VCore gets increase too since ratio should remain similar, but that's another story)
Since each motherboard/CPU combo differs it's impossible to test every combination, hence memory manufactures specify on which motherboards XMP profiles were tested on. If you use any other combination, you have to test it yourself and see at which command rate, timings and voltages you get stable results without exceeding safe voltages. And even if you are within limit but close at, you will most likely shorten the life of IMC considerably.
So, what are the benefits of running memory at higher command rate?
Increased memory bandwidth which looks good in some benchmarks but you don't really need it.
Slightly lower latency which is mostly negated by how internal memory caching works.
Better balanced system if you are significantly overclocking your CPU and GPU (mostly applies to extreme overclockers).
Slightly higher frame rate if you're using internal GPU (bellow 10% in average at 3600, and the reason why Intel likes to advertise XMP).
Higher speed rated memory has usually better silicon (product binning) but doesn't apply to all brands.
Negatives:
Higher $$
XMP does not guarantee stability unless you're using same components as memory manufacturer was using for testing, and even if you do, silicon is not silicon in many cases.
Almost no performance gain in real life usage. Gaming (with discrete GPU) performance increase is nearly non-existent. Maybe 1, 2 FPS, rarely more than that, most of the time less than a single frame.
Overclocking reduces the life of components.
If you really need to increase performance without shelling out extra $$, go for CPU and GPU overclocking. Performance gain will actually be measurable, and in most cases it's much safer because you'll probably hit thermal limits before voltage limits.
Memory overclocking is exactly the opposite. No real gain for a lot more $$.
