The way I currently understand things (by no means complete or authoritative!) is that there is a bit of a balancing act between share difficulty, ping time and block luck.

Not to state the obvious, but the lower the share difficulty the smaller a contribution each share makes to solving a block and absent any other factors, 100 shares at 100 difficulty is the same as 10 shares at 1000 difficulty or 1 share at 10000 difficulty. Obviously, lower difficulty shares take less time to solve on average - solving cryptographic puzzles is a statistical event, after all. I suspect this is well understood by most here.

However, the amount of time it takes to get work from the pool and send the solution (ie - share) back - usually just called "ping time" - starts to dominate the time it takes to generate each share as share difficulty goes lower. For example, if the ping time is 100ms and it takes 100ms to generate each share then your miner is sitting idle for 50% of the time, basically. This is perhaps not as well appreciated - it certainly wasn't by me for the first few months I was mining.

Finally, setting share difficulty too high can also result in a decline in net earnings if it takes longer to generate a share before the block is solved by the other miners. Something else which is perhaps not as widely or deeply appreciated, but I learned this the hard way with several cryptonight coins which all used the same broken vardiff algo that would ramp share difficulty up so high it would take my Ryzen 5 1600 15+ minutes to find a single share. I kept getting rejected shares and my earnings would hover around 50% of predicted. So now I will only mine CN (forks) if static difficulty is supported; other algos like, say, Equihash and Ethash, don't seem to suffer such wild swings in vardiff.

tl;dr - set static difficulty to get a share every 2 seconds to 2 minutes for best results.

I have been mining WTC WTCT  for the last week. It’s a very unique coin to mine. You essentially solo mine. I am averaging one block a day. The cards I am throwing at it are 6x 1080 ti.

Quick math that comes to $5 per day per 1080 ti.

The cards run very cool and quiet. I don’t have numbers but power usage has to be very very low.

To my knowledge this is a Windows only mineable coin at this time. Requires a beefy CPU as well.

thanks for the insight

2018/2019 will be the year of the return of the FPGA.
FPGA got kicked out when ASICs became maintream but were always used by some smart guys for very high profitability.
Some people have opened the pandora box.
Except if all major alt coins find a way to make their algorithm FPGA resistant (which is difficult), they will rule over GPU (high power consumption) and ASIC (bricks after a fork)

Personaly i have 4 rigs * 6 1070Ti plus one rig of 6 vega 56 and i will definitely make the switch.
You can have much higher efficiency but it's indeed less versatile than GPU.

GPU will stay the leaders for new coins using unusual and new algorithm.

You might be able to run one gpu with that. I am not sure.

Yes that is the coin.  When you mine a block I believe reward is 4.  GMN pool owner was keeping 1 coin for himself and giving you 3 coins. 

In order to have a GMN u need to have 5,000 WTC coins before December 2017.  So GMN exist now, my understanding is it's not possible for more GMN's to ever exist, only decrease.

It might be the year of the FPGA but keep in mind that unlike GPUs the supply for those high end FPGA is pretty slim.

Most likely it will be like the GPU market, constant out of stock or prices 2* above MSRP.

Yep - even mid-range FPGAs would soon run out of stock if miners started using them; there is nowhere near the manufacturing capacity for mid- to high-end FPGAs that there is for GPUs. I'm talking orders of magnitude of difference in annual volumes - it's a fantastic year if a manufacturer sells 1000 of a high end FPGA...

Also (in response to @melpheos) there is no such thing as an FPGA resistant algo. Any PoW computation a GPU can do an (appropriate) FPGA can do better. But developing the hardware description language (HDL) requires a fairly rarefied skillset that is not interchangeable with "normal" computer language programming (e.g. - C, ASM, etc.). The integrated development environments (IDEs) for FPGAs/ASICs like Quartus and Vivado are incredibly difficult to master as well. This basically means there are a lot fewer people in the world who are capable of programming an FPGA for mining, and of those that can it is still a more difficult task, relatively speaking, than writing a C program to do the same (not more difficult than writing an ASM program, however - you have to be nucking futs for that; or a masochist; or both).

Did you not read OGodAGirl article on this?
It is possible as you have to make the algo use as many parts of the GPU as possible. Then you can't strip down and just put an asic chip and memory. FPGA as good as it sounds will never make mainstream like GPUs. They haven't even had the 2x mining tax put on them yet and they are super expensive