If you buy a bare Pi you have a bit of work to do to have a complete general purpose computer. That's appropriate because many of them are sold for embedded systems rather than to be used that way. But people are more computer savvy now and most don't find it a challenge to gather the other parts; if you want something simpler you can buy a complete kit that has it all (Pi board, case, NOOBS card, power supply) in the package. Still a touch of assembly required: you have to put the NOOBS card in the board and the board in the case.
You also have to provide your own keyboard, mouse, and display. (Name brand desktop systems come with a keyboard and mouse, but the display is still a separate purchase unless you buy an all-in-one.) Many people already have those things left over from other projects, and if not your local office supply store or electronics store, or Amazon or eBay, will be happy to sell them to you.
Good point. Even those under-$100 B350 motherboards will take 64GB; all the ones I have seen have four sockets. I have seen pictures of A320 motherboards with two sockets as well as with four, but no A320 boards seem to be available in the real world yet. The small form factor A300 and X300 boards will be limited to two sockets and 32GB when they finally appear.
I doubt that AMD designed in any artificial roadblocks; it's not their style. So that ceiling will increase when larger DDR4 sticks become available. Right now 32GB sticks are only available as registered ECC modules and consumer motherboards don't accept those, but larger unbuffered sticks should be available in the future when higher capacity SDRAM chips are released.
Initially, these are probably all going to be eight core die with some cores disabled. But there are some interesting performance implications depending on what parts of the chip are disabled. It's entirely possible that different six and four core chips of the same model will not perform identically.
At the six core level there are two possible configurations: you could have one where one of the four cores of each complex is disabled, and another where one complex is fully enabled and the other has two cores turned off. Each type would require different process scheduling to perform optimally.
At the four core level you could have chips where one complex is turned off entirely, and also parts where both complexes are partly active. The low end Ryzen 5 1400 has only half as much L3 cache, so it's almost certainly the first type and runs only one complex. The 1500X has the full 16MB of L3 so it's presumably the second type. That could be either 2/2 or 3/1 and each would have to be scheduled differently.
GPUs also do this. A lot of the model differentiation in video cards involves selling of cards that are partially disabled and/or downclocked because they failed to meet specifications with everything turned on. They may have some completely broken parts, they may have failed at the full clock speed, or they may have consumed too much power at that speed. In some cases they actually passed all the qualifications but are sold as lesser (and less expensive) parts because of lack of demand for the most expensive model. When you buy there is no way to know which of those things you got, though you can run tests at home to try to figure it out.
It also happens with parts that don't contain processing power. DRAM comes in a variety of speed grades but they all come from the same fab line; they get sold with various speed ratings (and prices) based on how well they perform when tested.
Depends on your usage. The 1600X looks like a better gaming CPU than the 1700 and it's effectively $50 cheaper. (The list price is $80 less but no cooler is included.) On the other hand it will consume more power at peak; the 1600X is a 95W TDP processor while the 1700 is 65W TDP. We won't know what the idle power story is until the Ryzen 5 CPUs are released and benchmarked.
Looks like the i5-7600K and i7-7700K have some serious competition ahead. The AMD parts will cost less than Intel's, even after the recent cuts in street price of those CPUs, and AMD motherboards are also less expensive. By the time they're available, the scheduling issues that are holding back Ryzen's performance in games should be resolved.
The important differences between various generations of CPU don't affect real mode operation, the old school mode that is used to run MS-DOS. They come into play in kernel mode instructions that are used for things like protection, memory management, and multitasking. Some changes have to be made to Windows, as well as other operating systems such as macOS and Linux, every time a new CPU generation becomes available.
In some cases, the new CPU will work with code for older OSes but won't perform optimally. We know now, for example, that the disappointing performance of Ryzen in many games is due in part to the Windows scheduler not being properly optimized for the new CPU architecture. There are three things it does wrong. First, you want to schedule all the cores before you start to schedule the SMT threads; Windows already gets that right for Intel Hyperthreading but not yet for Ryzen's equivalent. Second, if you do schedule SMT threads you want to put threads of the same applications on both threads of a given core. Third, Ryzen has a split L3 cache architecture: cores 1-4 have direct access to one half of the cache and cores 5-8 have direct access to the other half, and cache access to the wrong half is much slower. For optimum performance, the scheduler needs to take that into account, keeping all the threads of an application on one side of the split whenever possible.
May Euell Gibbons eat your only copy of the manual!