K0rN b4LL

Apart from saving a bit on power, it can also make the PSU quieter. Potentially by quite a lot, but it also depends on the heatsinks and fan.
 
Going from eg. 80% efficiency to 90% efficiency only saves you 11% on your power bill, but it cuts the heat generated inside the power supply in half (imagine the difference for a GPU or CPU cooler if you cut power consumption in half).

size, get wht your case supports
 
You need both a 8 pin and 24 pin, every modern psu has this.
 
cheaper. Failure is uncommon and effiency difference is tiny.
 
Really just small power bill savings, probalby won't pay off. But normally higher 80+ is a just better built psu, but not alwys.
 
Whats youre system build?
 
YOur probably overthinking this. NOrmally just get a good 500-700w and your good.

Size and mounting way. Pick the one your case supports. Usually that will be ATX, but maybe SFX.
Well that is the difference. The connector and thus the power that goes through it.
The 24 pin (ATX) is also what is used to turn on the PSU and is general board power, while the EPS is CPU power.
They plug in here:
 
The benefit of modular is just ease of installation, or rather ease of cable management.
They don't tend to go bad, but you will have a warranty regardless.
The rating itself is not that important for quality, but only for power usage, but the difference won't be huge.
If you want more info, read this:
But the general gist (on 230V, 100% load): 
Bronze = 85% efficient
Gold = 89% efficient
Titanium = 94% efficient
Those won't be very large improvements, but may lower your power bill a couple of euros/dollars/whatever a year.
 
Don't choose a PSU based on efficiency though, pick a good one. 80+ is not an indication of quality, merely indication of efficiency. There are bad 80+ Gold units and good 80+ Bronze units  and vice-versa.
Most people won't have to worry about this at all, unless you're running weird configurations/heavy power draw components (like multiple GPU's or HEDT CPU's)

I agree with the above post. Been running an R9 390 for several years & I'd love to have Radeon's driver pack work Nvidia hardware. 
I like that AMD includes overclock, diagnostics, game profiles & their ReLive all in one package.
With Nvidia, you need the control panel (which I have to get through the Microsoft store because the DCH drivers don't include it), then GFE for game profiles & streaming/recording (and all those settings can only be accessed thru the overlay), then another app for overclock, and possibly one more for diagnostics (depending on which overclock app you got). 
And for some reason, with my new RTX 2060, I have a Display Configure icon in the taskbar showing me what apps are running on which GPU - yet I only have one GPU & one display. Not even an iGPU right now. But I can't get rid of it; option is greyed out.
 
So yeah, I'll take Radeon drivers any day. 

Their drivers are comparably stable to Nvidia's drivers. There were a few early issues due to using a completely new architecture in the 5700 series, but that has almost completely been fixed. (and in about the same amount of time it takes Nvidia to do the same with their new GPU architectures)
 
The "AMD drivers are shit" stuff is severely blown out of proportion, and not at all true.

If you can afford B die sticks then get anything with them. I don't know if B-die finder is accurate, a guess it is.

Clock speed and Timings are adjusted individually, so you can increase the number of clockspeed without touching the timings. Yyou can increase your memory clockspeed to lets say 3600mhz with the same timings, with this you have significant performance boost ( decreased latency) but often you need to increase voltage OR relax timings which are the number of cycles needed to perform a task like precharging memory or accessing a colomn or a row in the memory, or how much time the specific location on the memory is accessible to the CPu to read data. When you increase the clockspeed  you reduce the amount a time cycles last, so lets say you need to open a row and colomn in the memory for a total of 36 cycles, those cycles may last 100 nanoseconds but since you increase clockspeed by 15% this operation takes now 85 nanoseconds and it's not enough time for the data to be located. To counter this you relax the timings by settings those two actions to 20 cycles instead of 18 cycles, to give you back the 100 nanoseconds needed to perform those tasks.
Thats not exactly how it work, its way more complex then that but if you can understand the basics of that then you will have a greater time overlocking ram.
Thats why RAM performance need to be calculated in latency, because higher clockspeed with relaxed timings does not mean its faster then lower clockspeed and tighter timings.

They can change as you change frequency with XMP/DOCP enabled. At least this is what boards I've used so far do.
 
It's just like asking how much sugar is too much in coffee/tea in terms of numbers of crystals, really hard to tell. I'd say more than 2 ticks of change in the first number has noticeable effect in performance, while more than 1 tick of change in the first number, more than 2 ticks of change in the second and third number has noticeable effect in the kit's OC potential.
 
Dont give a s*** about that. My case does not have transparent side panels.

Usually the second and third timings ( tRCD and tRP) are the same. So you see something like 14-14-14-34 or 14-16-16-36. To make things simple, the lower timings the better, and something like 3200mhz 14-14-14-34 usally can be overclocked to 3600mhz and more with very tight timings. On the other hand, 3000mhz 16-18-18-38 is usually Hynix A-dies and they don't OC very well. I would pick tighter timings over fast clockspeed because you can loosen the tight timing a bit a get faster clocspeed, like getting a cheap Crucial Ballistix Sport 3200mhz 15-16-16-38 with Micron E chips to 3600mhz with 16-18-18-36, or even 3200mhz 14-14-14-34 is possible with good chips.
 

I would get this, you won't even notice a 5% increase going with a 275$ Trident Z kit, Crucial doesnt use cheap Hynix A dies so they generally OC very good.

Best value: 3200MHz CL16, since they are right before prices start jumping up really sharply. 
https://pcpartpicker.com/product/gLGxFT/crucial-ballistix-sport-lt-16-gb-2-x-8-gb-ddr4-3200-memory-bls2k8g4d32aesbk
Crucial seems to dump a lot of high quality memory dies here, which means cranking them to high frequency and still decent timings (say 3600MHz CL15) without too aggressive voltages is very possible.
 
best performance: on or slightly below 3800MHz with the lowest timings. Since you can't actually buy Gskill's 3800MHzx CL14 kit, the best goes to their 3600 CL14 kit
https://pcpartpicker.com/product/MFdrxr/gskill-trident-z-neo-16-gb-2-x-8-gb-ddr4-3600-memory-f4-3600c14d-16gtzn
Dabbing the voltage up a bit should let it run 3800MHz at even tighter timings no problem.

Or you can simply increase air speed to help draw out the heat more.  A good airflow case and a good air cooler will be fine.

I would get a large air cooler like the nh-d15 its better then all but 360mm aios and open loop, my ambient gets to 43C in summer and my house has no ac and yet i can still run 4.7ghz without going over 85C on my 8700k (no delid)  in cinebench with that ambeint.