DDR4 questions

Looked through all the reference specs and datasheets available, the wikipedia DDR4 article covers most of the basics well enough. Be aware that I consider myself a somewhat "expert" technerd on the engineering electronica for old SDRAM and DDR and even GDDR, but DDR2 (let alone DDR3) is just an entirely new beast which shakes my confidence - and DDR4 is just a big expensive bag of questions.

At this time I lack the variety of components which could show actual benchmark results and comparisons. I couldn't find anything online which isolated these details for useful comparison. I guess it's too new, the people with the toys tend to focus on overall brag rather than methodical component metrics.

1)
Which of these would be "faster" and "better"? And why, in what ways, for what tasks, by how much?
CL12 DDR4-2400 or CL15 DDR4-2600?

2)
Are larger memory modules "slower" than smaller memory modules rated for same base CL and MHz? Say, an 8GB module and a 4GB module, both CL15 DDR4-2400?

3)
Would the same memory run "faster" or "better" in a motherboard capable of supporting higher base (that is, not-overclocked) frequencies? Say, a quad of 4GB DDR4-2400 modules in a 3200MHz-capable mobo vs a 3333MHz-capable mobo?
What exactly defines a mobo's maximum rated memory frequency, anyhow? The manufacturer's particular choice of chipset implementations and mods and tradeoffs, the bandwidth capabilities of some particular onboard address-bus part, the impossible-to-calculate overall gestalt of copper-trace lengths/geometries and conductivity characteristics of the paint in the fancy brand logo, the marketing team, the phase of the moon?

4)
I understand well the general quirks involved in overclocking, component tolerances, variable qualities of individual pieces from same mass-produced batches, etc. Is a "quad-channel DDR4 kit" - supposedly four "identically matched" DDR4 modules - actually a better guarantee (or at least a better gamble) than four individually selected DDR4 modules with identical spec purchased over time?
I guess I'm really asking how far a memory manufacturer actually goes toward matching the modules. Just four of the same thing from (hopefully) the same batch run bundled together a bit cheaper? Or four exhaustively tested and meticulously "balanced" pieces selected for optimum functionality? Does this vary from manufacturer to manufacturer, brand to brand?

5)
DDR4 spec supports a few extra pins dedicated to module-integrated microcapacitor/clock/logic/NVRAM circuitry - the purpose is to preserve memory contents during momentary power loss (perhaps even long-term power loss), an appealing feature for mission-critical server type machinery, and maybe - I suspect - even another buffer to push overclock thresholds. DDR4 modules can (eventually) be purchased with or without this component (whatever they gonna call it), and these can be used interchangeably in the same sockets and platforms. Or so says the tech literature.
My concern here, as usual, is the presence of NVRAM (rewritable-flash ROM stuff) in a component/device which is expected to see countless continuous writes and rewrites as part of its normal operation. Should I expect a DDR4 module to "burn out" after a year or three just because this one redundant subcomponent has finally exceeded its capacity? Could such a "burned out" module be resurrected by just flipping a DIP, replacing a socketed part, or practicing some home dentristry on a few pins? Anyone know more about this, in terms of DDR4 implementations?

(On a somewhat related note - hopefully not gonna derail my own topic right in the OP - mechanical HDD media are typically rated in terms of "MTBF", flashy SDD media are typically rated in terms of "maximum number of write-cycles" ... can anyone linky sites which provide real-world data and head-to-head comparisons of expected media lifespans? Hell, even statistical data is better than no data at all.)

I'll start with: DDR4 RAM is out of my depth.

I can help you with the little tiny question in brackets at the end of the post though. The Tech Report decided to see when a bunch of consumer SSDs would die if you wrote data to them until failure. Pretty interesting results over the 1+ year it took them to run the experiment. Hardware.info did some lifespan testing on the Samsung 840 when TLC NAND was hitting the market which also make for some interesting reading. Long story short, it's not the number of writes to an SSD that kills it in a consumer setting. My guess, it's probably issues with power. A good PSU behind a decent line-active UPS and most SSDs should likely last as long as most HDDs.

Hmm, thanx for the clues and linkies, sir. Exact cause of death isn't as significant as the fact that death is inevitable, nothing lasts forever. I'm still a pure HDD kinda guy, I've seen many such drives function perfectly for decades, and I'm still apprehensive about devices which (entirely due to their vastly superior performance) can approach their finite limits that much faster.

Nonetheless, it's time to move along. The raw awesome sauce offered by a performance SSD RAID pair is just too excessive to pass up. To answer my own question (5) above, it seems the extra reference pins are an entirely isolated and self-powered subsystem which can be extracted from the memory module without diminishing anything (other than deactivating their own NVDIMM functionality, of course).

Alas, question (1) is the most immediate one on my mind at the moment. The almost invariant rule back in the pioneering DDR (er, "DDR1") era was that lower CAS latencies and higher MHz just went together. But semiconductors and embedded logic have changed dramatically since then, I honestly don't seem to know anything relevant to modern computer hardware, lol.