But more important than everything else: this mustn't turn into a sacred war. It's something for grown up men discussing like good friends drinking beer in a bar. 
If it's something to bring up anger, frustration and ferocious discussions, I rather leave the MSX3 as a failed chapter of the past and keep my good friendships here. No MSX3 is worth more than the 25+ years old friendships I have here.
IMO, we need to converge rather than diverge. Also, make things easier to develop and debug with less bottlenecks but keeping the compatibility.
So, IMHO, the effort can be focused in building V9990+V9958 cycle model and R800. There is no problem with sourcing VDPs, but DAR800 is absolutely impossible to find now days. So at least R800 must be repeated in the HW.
But again, it is about keeping the current MSX2+/TR architecture and removing the bottlenecks, IMHO.
Unfortunately all of the new MSX hardware (OCM, GR8BIT) is built on MSX2/2+ specs, and since turboR computers are expensive and relatively hard to get, I hesitate to fully utilise them, never mind require one.
I endorse nearly everything you said, but as someone who already dealt within the hardware development world I must add this reality check:
Given the constraints, a new MSX3 wouldn't be exactly cheap to produce. Those constraints are:
1) Small scale, mostly hand made production.
2) Obsolete chips that aren't cheap or easy to obtain
3) The highest cost wouldn't be the OPL4, but the compatibility costs accumulated to the hardware side because of software that didn't follow the guidelines. Please don't get me wrong, I'm not saying this as a "code according to the guidelines" campaign. It's what I would call "the MegaFlashROM SCC+ SD effect ": there are people that complain that it's too pricey for a cartridge. But that small cartridge has to pack a lot of hardware power to be able to be compatible with a plethora of different MegaROM controllers, plus SCC+/PSG and SD cards. And I can assure anyone that a MSX3 would cost a lot more than the MegaFlashROM SCC+ SD 512KB just because the hardware would have the burden to be compatible with *a lot* more software.
Corollary of (3): There will be that point in the design road where we will have to decide where the compatibility costs will be pushed to. Maybe to the hardware side (very expensive MSX3 with a monster try-to-save-the-world FPGA) or rather to the software side (patch existing software to behave according to the coding guidelines).
Or... just give up a and forget all about the MSX3 thingy. 
And... Depending on what choices you made, it will be or not possible to enable nifty things like DMA & cache. But forget about DMA & cache with badly behaving programs.
3) The highest cost wouldn't be the OPL4, but the compatibility costs accumulated to the hardware side because of software that didn't follow the guidelines. Please don't get me wrong, I'm not saying this as a "code according to the guidelines" campaign. It's what I would call "the MegaFlashROM SCC+ SD effect ": there are people that complain that it's too pricey for a cartridge. But that small cartridge has to pack a lot of hardware power to be able to be compatible with a plethora of different MegaROM controllers, plus SCC+/PSG and SD cards. And I can assure anyone that a MSX3 would cost a lot more than the MegaFlashROM SCC+ SD 512KB just because the hardware would have the burden to be compatible with *a lot* more software.
I think for the as-of-yet imaginary “MSX-ENGINE-3” the FPGA size used in the 1chipMSX should be more than sufficient, because although it adds an R800 and S1990 it does not need to include a VDP or OPLL. That FPGA is quite old and low capacity. If the MSX turboR+ can be made for a price similar to the Zemmix Neo (and in an enclosure just as cool ), I’d be more than happy…
An FPGA of that size costs around 30-40 euros… For a cartridge that indeed adds a lot to the base cost, but then again the MegaFlashROM SCC+ SD is basically an all-in-one MSX to 1chipMSX upgrade cartridge, so yeah, if people want something cheaper then they shouldn’t buy something which does so many different things at once .
I think if in stead of an FPGA you would combine a discrete Z80 and PSG with a CPLD or even discrete logic ICs (and a gigantic board), I don’t think it would be much cheaper, if at all…
Corollary of (3): There will be that point in the design road where we will have to decide where the compatibility costs will be pushed to. Maybe to the hardware side (very expensive MSX3 with a monster try-to-save-the-world FPGA) or rather to the software side (patch existing software to behave according to the coding guidelines).
I think you’re forgetting one cost: that of the developers who are going to create all those software patches in stead of creating software that is actually interesting, and the cost of the user having to deal with all those patches. And people not playing a great game because either nobody patched it, or the website which hosted the patch went down, or because it’s just too much trouble, that’s also quite “costly” to the MSX heritage if you ask me.
Personally I’m not interested in an incompatible MSX turboR+, if it requires me to patch many of my software. I’d rather pay even €50 more and have my life made much easier…
On the other hand there are a few ppl making new msx boards though, so the skill is out there. And FGPA (which is not emulation) makes it a bit easier /accessible to make new hw.
Exactly.
In the mass production, FPGA is the only solution potentially providing the blueprint functionality of the original HW without a burden of finding the obsolete parts.
The quality of the FPGA replica is up to the designers, their motivation and available spare time to dig into every tiny detail of the original design. And this is time consuming.
One more argument is that the "good" FPGA design can be retargetable to any other FPGA technology or potentially an ASIC.
Also we shouldn't forget that any complex IC design 40-30 years ago was preceded by the huge board full of TTL logic, which was running the ASIC model. It was called "the sign-off emulator".
So, now, 30 years later, in FPGA we are actually trying to reproduce this model through reading the specs, reverse engineering by running the tests on the original HW. And this is where the main investment of time goes.
Ideally, IMHO, in the MSX community, we can split the problem on pieces and work on the precise reverse engineering of each IC in question. Someone could help in the specific test development (Zexall for VDP/Audio etc), someone can contribute the precise timing diagrams/traces to simplify the reverse engineering. This consolidation of efforts will earn very deep understanding of how the HW functions, which is the key to any replication.
Personally I’m not interested in an incompatible MSX turboR+, if it requires me to patch many of my software. I’d rather pay even €50 more and have my life made much easier…
Me neither. POKE -1, XXX is already bad enough across even MSX compatible machines
before embarking on the MSX3 HW definition
It is empiricaly proofen that talking a spec ends up nowhere
To make anything new, this needs META talk.
Those interested in a new machine, what do they already have?
Is it the usual MSX2? That has no OPL, so you gotta have a cartridge. But instead just plugging it you say "I wont buy that MSX3, it lacks opl soandso", is it that?
The part that you actualy need to take care of is the MSX3-ready logo. Which means it has a fast z80 plus the 9958 wait pin working. That's the basis for all that will be coming.
The same machine I just talked is also the one for those who always wanted a TurboR.
From some random thread: "I have not seen a TR for sale below €350 in a very very long time".
In the other thread was posted about a discrete MSX with surprising low price, I always thought it is factors more than the little smd fpga board.
The OCM has the infinite feature list. There is no demand for it? Why?
