What Do You Think Will Happen Once MiSTer Reaches It's Limits?

[Nodoyuna]

Hi

MiSTer is great, no doubt about that. There are many cores available and there are many more in the pipeline

Anyway, sometime the MiSTer will reach its limit and more complex cores simply will not fit in the FPGA

There is some kind of consensus about the MiSTer limit will reach with the PS1 and/or Saturn cores, so the MiSTer will become a plattform for all 8 and 16 bits machines and maybe (if PS1 and Saturn gets done), 32 bits too

I was wondering about what will be done then, I've thought of a couple options:

1) Keep MisTer for 8/16/32 bit cores and move to a new more powerful FPGA for more complex machines only (Dreamcast, PS2...)
2) Get a new more powerful FPGA and conver all MiSTer cores to the new FPGA (similar with the MiST -> MiSTer "transfer")

I'm curious about what do you think will be the next step

Cheers

[FoxbatStargazer]

The next step is companies releasing more affordable FPGAs. From what I understand its a huge price leap to something more powerful than the DE-10. If companies don't feel like doing that then I'm not sure we see movement. You also got the problem that more powerful systems are significantly more complex and less likely to ever be finished.

So the likely order of what will happen, if anything happens:
1) Significantly stronger FPGA becomes cheaper (who knows when)
2) Someone releases an awesome core or two not possible on mister. (lots of work so again, who knows when)
3) Now that new cores are drumming up interest in the board, mister conversions start happening en masse.

I think the only time most Mister cores get left behind is if the new system uses some really different method. Like maybe having an actual GPU instead of trying to recreate those in FPGA.

[Sorgelig]

There are limits where FPGA can be used regardless if it's FPGA in MiSTer or any other boards.
PS2, Dreamcast and later consoles/computers use very high clocks not reachable on FPGA. Also chips became more complex and less low-level technical info is available. To create the system in FPGA you need much more technical low-level info than for software emulator. And more modern system - more hard to find such info.
FPGA will remain at the 90x systems emulation in foreseeing future.
Also there is no real benefit of FPGA against software emulators on later systems. They are 3D systems where GPU is more useful than FPGA. Also these systems aren't depend on cycle accuracy.
So FPGA is pretty much fixed for pre-2000 systems.

More modern FPGAs may give some advantage for later computer systems such as PC or Amiga. Saturn is on very early stage, and PS1 is in unclear state too - so if they will fit into MiSTer then it will be pretty much the top of possible.

FPGA won't be better on PS2 emulation or anything more modern.

[dshadoff]

I think that once MiSTer reaches its limit, some of the developers will get to play some games.
Seriously though, there are always tweaks and improvements which will be researched and implemented, and still hundreds (thousands ?) of arcade boards to implement.

[Lightwave]

I forsee a long and bright future for Mister, even just based on arcades alone (as dshadoff mentioned).

Tech specs and new technology aren't necessarily the main factor for a platform. The real success of MiSTer comes from the combination of its maintainer (Sorgelig), skilled developers offering their time (numerous) and the enthusiastic user base (community support). We can't assume that the magic combination of these things can automatically be replicated with a new project/platform.

The other thing is, even if is MiSTer is ultimately restricted to 16/32 bit machines and lower, the fact that ultimately we will have all of these consoles, computers, and arcade games preserved accurately (in both function and playability) on a single small board, is a miracle. If this was a product sold in stores with a nice case and supported/promoted by major retailers and game companies, I'm sure it would become one of the most popular and bestselling gaming/computing devices of all time.

Personally, I can't imagine ever giving up my MiSTer, even with the advent of some theoretically more advanced new platform. MiSTer already performs pretty much perfectly for everything I use it for (and I have no doubt any existing issues with accuracy will be corrected). In the same way people keep original hardware because it just works, MiSTer should always have a place in the future for any gaming/retro-computing enthusiast, even as secondary system or just something to tinker with.

[aberu]

There are limits where FPGA can be used regardless if it's FPGA in MiSTer or any other boards.
PS2, Dreamcast and later consoles/computers use very high clocks not reachable on FPGA.

I agree that PS2 is likely not going to happen in the near future (10+ years) on FPGA, but this is mostly due to how notoriously complex the PS2 is and how little of it has still been actually reverse engineered. Emulators require complex plugins to be loaded and parameters changed by the user just to get each game to run close to correct. Most games in PS2, especially later games after developers learned how to develop for it later on, have significant issues that are noticeable in emulation to this day. An FPGA PS2 would be exponentially more complex than a Saturn FPGA implementation, which is already difficult.

As far as Dreamcast it's actually a fairly simple system by comparison to the PS2, it's just more powerful. The main thing is bandwidth for the processors and the memory. It could be done in FPGA, but at what cost? Someone could probably do the Dreamcast in a $20-60k FPGA PCIe board today if they spent many many years on it . But that's unreasonable to expect anyone to do, and no one would buy such an FPGA board just to play Dreamcast on it.

But in terms of the regular MiSTer user and target audience, I agree that Dreamcast will probably not be seen in the next 10+ years in FPGA. Too much parallelization and multiplying needed for that kind of bandwidth from how little I understand it. And that sacrifices timing when you do too much of that as a crutch.

Glad to see my assumptions from my amateur perspective were mostly correct.

But much higher bandwidth than any of the current cores on the MiSTer absolutely can be achieved, but they come at the cost of taking significant space on the FPGA to the point where it won't do much of anything else than just be a GPU, or a fast CPU.

https://www.researchgate.net/profile/Al ... ion_detail

Here's an example of someone doing equivalent GPU performance to an Nvidia Tegra K1 in terms of vertices per second at around 1% of the power consumption, on a Cyclone V. He achieved 197-210MHz on the Cyclone V with this study. So I think it's more accurate to say that it's unreasonable and that there isn't enough space on any FPGA available to a regular consumer to emulate these whole systems. Just saying that these FPGA can't achieve this speed isn't taking into account the potential for a multiplexor + highly parallelized graphics pipeline to achieve much higher bandwidth. However, again, I want to reiterate that the limitation of space, maintaining timings when doing this, etc... becomes the real challenge and the likely limiter of what is possible.

2021-04-27_9-43-43.png (73.77 KiB) Viewed 30895 times

2021-04-27_9-44-19.png (72.74 KiB) Viewed 30895 times

2021-04-27_9-44-53.png (58.96 KiB) Viewed 30895 times

2021-04-27_9-45-43.png (362.45 KiB) Viewed 30895 times

2021-04-27_9-46-22.png (123.57 KiB) Viewed 30895 times

2021-04-27_9-47-30.png (168.46 KiB) Viewed 30895 times

[Neocaron]

Can we expect the hybrid approach with CPU emulation running on the arm chip and the FPGA used for the rest to push the DE-10 a bit further? The first test on this look pretty promising. My hope wouldn't be the Dreamcast, but more like 1996/99 PC and the N64 with CPU emulation running on the arm and GPU and other stuff on the FPGA. Tha would be pretty awesome to have a voodoo PC thanks to this.

[Bas]

If the ceiling is hard, there is still plenty of work to be done on far less obvious cores than what's being worked on right now. As a computer guy myself I'd say the C64, Speccy, Amiga and ST for instance were must haves to cover the 80's in any sort of meaningful way. But there were also the Apple II, BBC Micro, MSX and IBM compatibles. All of these have appeared. So are we done? Sure, the bases are covered.

But what about an accurate PC Jr. or Tandy 1000? A more accurate 386DX PC would even be of interest to me personally. There's also still considerable headroom between what the MacPlus core does, and what the DE10 Nano is capable of. There's also a fair bit of computer history worth recreating, like DEC VAX, older gear from Sun, SGI, or mainframes like the Burroughs large systems platform. Heck, the Apple III would be nice, as well as Lisa. These are obviously not gaming systems, but they sure are interesting to old CS geeks like me and they could fill conservation niches.

All of these would fit the MiSTer in its current form and would be appreciated greatly. And I'm not even mentioning the probably still many outstanding arcade titles not yet covered or more or less hardware variants to existing cores like the C128 or C65. Not immediately interesting and other work takes preference, but eventually MiSTer will encounter the "long tail" of diminishing returns. That's where computing history will just keep on giving new opportunities for additions to the platform.

[metalfacemark]

Can we expect the hybrid approach with CPU emulation running on the arm chip and the FPGA used for the rest to push the DE-10 a bit further? The first test on this look pretty promising. My hope wouldn't be the Dreamcast, but more like 1996/99 PC and the N64 with CPU emulation running on the arm and GPU and other stuff on the FPGA. Tha would be pretty awesome to have a voodoo PC thanks to this.

This would be fantastic. I also think personally that it lends itself much better to arcade and 8/16 bit stuff, keeps everything neat and tidy. If we can get working playstation and saturn id be more than a happy camper.

Dreamcast and Gamecube are easily done on original consoles with cheap gdemu replacements and nintendont with the wii and wii u. Xbox, PS2 can also be softmodded and the consoles are cheap to pick up. (although i dont bother with the xbox and have a sizeable ps2 collection so i havent looked into this).

[MiSTer_Kirk]

I'm more interested in cores for hardware that is unobtainable, or stupidly rare and expensive, rather than easily obtainable hardware (Dreamcast, Xbox, PS2, Gamecube).
Such as Cantab's Jupiter Ace, Memotech's MTX 512, Enterprise 64/128, Original IBM 8086 PC, Full spec 90s PC with hardware acceleration - a full set is now starting to become valuable on eBay, Lots of Apple machines still without cores, Then there are very rare, or even unreleased hardware such as the Nuon, or a working core taken from the only one emulator of a Konix Multi-System.
Then there are hundreds, or even thousands of Arcade boards that still don't have a core, especially from the late 70s, and early 80s.
There are cores for systems that had disk controllers, yet their cores still rely on the awful MMC method, such as the BBC Micro/Master.

Oh, I don't think we'll see the end of the Mister for a long time yet.

[dshadoff]

...And there is also still the other end of the spectrum to explore:
- super-old computers (WANG, Data General, etc.)
- obscure early home computers (Sol 20, Exidy Sorceror, Cosmac Elf, etc.)
- programmable calculators
- handheld games like "Mattel Basketball", game & watch, etc.
- possibly cores which visualize the internals of simpler systems

...and so on. The range is endless.

[Nodoyuna]

Hi
Thanks everybody for their opinions. There are some very interesting ones
Of course the future is endless for MiSTer, as there still are many many machines that can be run on it.
I only hope there will be interest from developers to make cores for more obscure machines
I also agree about more modern systems. They're not really needed as they are easily available and not very expensive
Please, keep posting your opinions
Thanks

[Kalvan]

Well, we could start looking at playing what-if, speculating about computers and consoles that coulda-woulda-shoulda been, if only the stars had aligned and the suits had had a lick of honor. For example, extending the Minimig architecture with the functionality of the planned Ranger Chipset that the suits of CBM shelved because they wanted to pinch every penny possible, letting the Amiga fall behind.

Another idea would be taking the Atari Lynx, replacing the 65C02 core with a Motorola 68000 core clocked at 16 Mhz, doubling up on MICKEY, widening the video interface to at least 8 bits (9 would be even better), adding a Yemeha YM2612 or YM3438, and quadrupling system RAM to 256K, and quadrupling the resolution to 320x204. Voila: A would-be Neo Geo Pocket system worthy of its elder arcade hardware and home console brethren!

My personal passion project would be to create in FPGA form a custom system with the graphics chipset based on the Atari Advanced Engineering Division projects (specifically, mostly based on the OMNI chipset, but with the addition of SILVER from the RAINBOW chipset, using a MIPS R2000/3000 or R6000 core for the CPU, with AMY, QuadPOKEY, and a couple other sound chips having their own Sound CPU and dedicated audio RAM.

If anyone wants to help out with brainstorming of this idea, please either respond, quoting this post, like it, or PM me.

[LamerDeluxe]

I see lots of cool ideas. The Cray super computers were mentioned as being possible, that would be really interesting.

I agree that a what-if logical Lynx successor would also be cool, though it wouldn't have software.

And the Amiga AAA chip set would be really cool. I have had a hard copy of the whitepapers since the early nineties, but I don't know if it contains enough information for creating a core. But again, it wouldn't have any software and the OS would have to be heavily adapted/made for it.

[dmckean]

More modern FPGAs may give some advantage for later computer systems such as PC or Amiga. Saturn is on very early stage, and PS1 is in unclear state too - so if they will fit into MiSTer then it will be pretty much the top of possible.

Nintendo 64 and Nintendo DS are two systems that are possible on FPGA but not on the DE-10 NANO. If the list got longer and included Saturn, PSX and some arcade boards I think that investigating more powerful FPGAs would be worth while. Faster PC and Amiga cores would make a lot of people happy judging by what I read here. It's probably a good idea to be investigating alternatives in the background anyway, just in case the DE-10 NANO suddenly went away.

[throAU]

In line with what Sorgelig said, once you get to say... PS3/Dreamcast era things are much less "direct hardware" and timing dependent and much easier to emulate in software via powerful enough hardware. Maybe PS2 as I hear that was a bitch to program, but it still had a PC-like discrete GPU which I doubt was synchronous with the CPUs.

You're dealing much less with precise timings that have all of the components in sync with one another, and more with a device using layers of abstraction to put work into queues that are run asynchronously - like say, a PC. Once you have things running asynchronously - you can split them off to multiple cores/threads in software and synchronised timing becomes so much less important. This is why, for example you can pull your PC's GPU out and upgrade it without breaking everything in your system - things aren't running synchronously to the degree they are in an 80s console!

The very precise hardware timing requirements of the old school 80s-90s era hardware simply aren't a thing beyond a certain point as the hardware is no longer relying on timing hacks for things like colour display output games, "racing the beam" for special effects, etc. The hardware simply became powerful enough to essentially "do it without hardware hacks".


As to trying to do Dreamcast in FPGA, I'd guess it is pointless. You could even program the thing with DirectX/Direct3d from memory - programming it was abstracted enough from the hardware that emulating it in software (accurately!) will be much easier than an 80s console.

[throAU]

This would be fantastic. I also think personally that it lends itself much better to arcade and 8/16 bit stuff, keeps everything neat and tidy. If we can get working playstation and saturn id be more than a happy camper.

Dreamcast and Gamecube are easily done on original consoles with cheap gdemu replacements and nintendont with the wii and wii u. Xbox, PS2 can also be softmodded and the consoles are cheap to pick up. (although i dont bother with the xbox and have a sizeable ps2 collection so i havent looked into this).

The problem you have with arcade and 8/16 bit stuff is that they are heavily timing dependent as above. Hybrid emulation won't work here very well due to the lack of the tightly integrated timings between components that FPGA provides.

The amiga hybrid emulation thing is feasible because one you go to CPUs on the amiga beyond the 68000 which the amiga supported (and typically this is for use with application software - not games), the precise timing requirements between the CPU and the custom chips goes away. This simply isn't the case for the old 8 bit and arcade hardware.

Ironically, the older the hardware, often the more hardware timing hackery was involved in the original design to make it work, and this often makes them more dependent on FPGA precise timing (than later hardware) to be accurate.

[ZigZag]

Same thing as happened when the C64 hit it's limits... Oh wait, it hasn't yet, and never will!

No reason why people can't refine cores & push the h/w limit's for decades. I understand it's not going to emulate an Intel I7 x64 PC, but there's still a virtually infinite amount of projects that could be undertaken within the h/w constraints of the MiSTer.

Sure, newer more capable FPGA h/w will become available at some point, but that doesn't mean people abandon & discard their old h/w completely. Surely the very concept of the MiSTer is irrefutable proof of this!

[Lodovik]

Is there a possibility to combine multiple DE10 boards together when the limits of a single board are reached or is it too complex to be practical or feasible?

[grizzly]

Is there a possibility to combine multiple DE10 boards together when the limits of a single board are reached or is it too complex to be practical or feasible?

You will have too use the DE-10´s io pins to tie them together and that will mean three things.
1, Speed!
We have trouble with the memory speed now and is practicably at what could be considered max speed.
2, There are not that many pins that can be used and for tying together two FPGA´s in any meaninful way there is probably a need for MANY.
And as it is we use many already for the memory/SNAC/vga/sound/optical, so not many left at all.
3, My guess is either tie FPGA´s together and run with no added memory (IF the pins will be enough EVEN then) or run a single FPGA with memory.

[aberu]

the precise timing requirements between the CPU and the custom chips goes away.

To a degree, yes. But timing of something like say... a math coprocessor in a Pentium chip is still important with gaming. It's still measured in nanoseconds. Additionally, it's a little misleading to suggest the problems go away after 8/16-bit, timing on the Sony Playstation and Sega Saturn are ridiculously important and difficult to achieve. Ask Rama about how hard it was to lock down the xStation over time, ask Nicolas Noble about all the intricacies of the Playstation, and look at how long it's taken to get a somewhat timing-accurate Sega Saturn with the modern Mednafen (somewhat, because tons of things are not timing accurate still). It's only recent. Now in FPGA it's "easier" to get cycle accuracy supposedly when compared to something like C/C++, but still...

[metalfacemark]

I see lots of cool ideas. The Cray super computers were mentioned as being possible, that would be really interesting.

I agree that a what-if logical Lynx successor would also be cool, though it wouldn't have software.

And the Amiga AAA chip set would be really cool. I have had a hard copy of the whitepapers since the early nineties, but I don't know if it contains enough information for creating a core. But again, it wouldn't have any software and the OS would have to be heavily adapted/made for it.

Doesnt that sound wild? As a kid the cray supercomputer always sounded like something out of wargames

[throAU]

To a degree, yes. But timing of something like say... a math coprocessor in a Pentium chip is still important with gaming. It's still measured in nanoseconds.

Sure, its measured in nanoseconds, but my point is it doesn't have to be exactly X nanoseconds.

Software written for the pentium will still run on a pentium 2/3/4 and all the way up to an i9 or Ryzen mostly.

Because the timing doesn't need to be exact, it just needs to be "fast enough", and the CPU for example doesn't necessarily have to run at the exact same timing as say, display adapter. Even amongst period-exact PCs of the day for example, RAM speeds, disks speeds, bus speeds, display adapter speeds and CPU speeds differed. You didn't usually run into software that ran fine on say a p90 but broke on a p166 for example.

That's my point vs. the old 16 bit and earlier systems, which have intricate timing sync requirements between individual components in the device. Not so much that they need to be fast or whatever, they need to be synchronised. PCs (and later consoles with a software development methodology via high level languages and hardware abstracted APIs) don't.

Emulating things fast enough is a simple problem of throw horsepower at it. Interdependencies with timing isn't so simple and that's where FPGA shines, because you can run different things synchronously and have the timing between the different elements line up with each other precisely.

[aberu]

Yeah I agree mostly with what you are saying, just that for gaming timing can still be thrown off and make things crash or behave strangely in the 32-bit era as well.

[justaguy]

If Dreamcast and up wouldn't really benefit from FPGA emulation, would a hypothetical "MiSTer but for more advanced consoles" really need to be FPGA-based? I love the MiSTer user experience for reasons that have nothing to do with it running on an FPGA, and I'd welcome a more traditional SD card image-based emulation frontend that focused on a core of generally-useful features and really took advantage of a standardized hardware platform. Something like RetroPie, but not trying to be all things to all people.

[aberu]

If Dreamcast and up wouldn't really benefit from FPGA emulation, would a hypothetical "MiSTer but for more advanced consoles" really need to be FPGA-based? I love the MiSTer user experience for reasons that have nothing to do with it running on an FPGA, and I'd welcome a more traditional SD card image-based emulation frontend that focused on a core of generally-useful features and really took advantage of a standardized hardware platform. Something like RetroPie, but not trying to be all things to all people.

Exactly right. I think for those systems and later a "baremetal emulator" as they can be called would probably be the optimal way to achieve low input latency emulation of these systems.

[FoxbatStargazer]

I'm hearing that Dreamcast et. al. aren't really feasible in FPGA but I'm not convinced that software emulation is somehow less limited than it is for old systems. There are some great Dreamcast emulators but even run on a full x86 with an adaptive sync gaming monitor, the input lag is pretty high side-by-side with a real system. You pivot Sonic around on that dreamcast controller and the response is just absolutely instantaneous and smooth. Same with a Gamecube vs Dolphin. Maybe there is no feasible solution for this but it still seems like something is missing with software emulation, which is why we got interested in FPGA for older systems in the first place.

[LamerDeluxe]

I see lots of cool ideas. The Cray super computers were mentioned as being possible, that would be really interesting.

I agree that a what-if logical Lynx successor would also be cool, though it wouldn't have software.

And the Amiga AAA chip set would be really cool. I have had a hard copy of the whitepapers since the early nineties, but I don't know if it contains enough information for creating a core. But again, it wouldn't have any software and the OS would have to be heavily adapted/made for it.

Doesnt that sound wild? As a kid the cray supercomputer always sounded like something out of wargames

Right? They sold for about ten million at the time. I saw them mentioned in documentaries about 3D animation. The stuff of dreams. The art school I attended had a deal to be able to use one at one point, sadly I never got to use it.

[aberu]

I'm hearing that Dreamcast et. al. aren't really feasible in FPGA but I'm not convinced that software emulation is somehow less limited than it is for old systems. There are some great Dreamcast emulators but even run on a full x86 with an adaptive sync gaming monitor, the input lag is pretty high side-by-side with a real system. You pivot Sonic around on that dreamcast controller and the response is just absolutely instantaneous and smooth. Same with a Gamecube vs Dolphin. Maybe there is no feasible solution for this but it still seems like something is missing with software emulation, which is why we got interested in FPGA for older systems in the first place.

I think that the term limited is not limited to just input lag as a topic. Upscaling resolution of the 3d models, enhancing texture filtering, etc... there are tons of things that software emulation will always do better than hardware emulation with FPGA.

Each of them have their strengths and weaknesses. We are just now getting save states in some cores, whereas software emulation has had saves states in almost every emulator for over 2 decades.

[dmckean]

I'm hearing that Dreamcast et. al. aren't really feasible in FPGA but I'm not convinced that software emulation is somehow less limited than it is for old systems. There are some great Dreamcast emulators but even run on a full x86 with an adaptive sync gaming monitor, the input lag is pretty high side-by-side with a real system. You pivot Sonic around on that dreamcast controller and the response is just absolutely instantaneous and smooth. Same with a Gamecube vs Dolphin. Maybe there is no feasible solution for this but it still seems like something is missing with software emulation, which is why we got interested in FPGA for older systems in the first place.

Writing baremetal emulators with the same delay as the original hardware would be no easy task either, but it's a couple orders of magnitude more feasible than implementing complex hardware with a hundred million transistors in FPGA.