PC DDR5 memory latency specs

[bson]

I'm in the process of refreshing my now somewhat ancient PC workhorse, built in 2015...

Looking at DDR5 memory I see specs like 6400 MT/s latency 30-40-40-72 and such.  This leads to a bunch of questions I thought you guys here might be able to answer... Are these latencies the number of clock cycles at the max specified transaction rate?  Or some standard JEDEC (e.g. 5600) rate?  In other words, given two sticks at say 6000 MT/s and 7200 MT/s with the same latency cycle counts, the latter would have slightly over 1/6 lower latencies?  Also, does the transaction spec take into account boundary crossing latencies for linear access, or is it the max specified transaction rate for up to one perfectly aligned page, ignoring the latency to initially switch to that page?  Do all PC DDR5 sticks have the same page size?

[golden_labels]

Memory timings are given at stick’s rated clock frequency. Which is why they are so huge nowadays, compared to single-digit values of the past.

DDR5-6400 operates at 3200 MHz. A single cycle takes 1 / 3200 Mhz = 312.5 ps. Timings 30-40-40-72 correspond to: 9.38 ns, 12.5 ns, 12.5 ns, and 22.5 ns.

Yes, the one with higher rate and same timing numbers will have lower absolute latency.

For the second part: I don’t know.

[magic]

The MT/s rating is simply twice the rated clock frequency, so you won't get that many actual data transfers in practice.

Dunno about row size (is "page" a new thing in DDR5, or is "row" what you really meant?) but in earlier generations most sticks seemed to stick to the maximum supported by common CPUs. That being said, AMD supported one bit longer column addresses than Intel and I recall there being DDR2 generation sticks which worked on AMD but not on Intel and they could have up to twice the capacity of common sticks. I'm sure the DDR5 DIMM spec also imposes some limit on maximum column address.