MiSTer FPGA Forum

I think discounting the idea of a dedicated retro gaming FPGA board is a bit silly, MiST is a great example of what such a board can do, it has a FPGA with only a quarter of the logic elements as the Cyclone V, but it's capable of simulating a majority of MiSTer's platforms. A MiSTer successor board can be a custom design that trades the DE10-nano's (evaporating) affordability for ensuring none of MiSTer's weaknesses or need for daughter boards. It will probably end up being more of an investment than MiSTer, but an ITX-style board that can simulate up to Gen 6 or 7 consoles and handhelds will only become more desirable as original hardware gets more expensive and scarce.

softtest9 wrote: ↑Wed Dec 01, 2021 1:16 pm You can load cores from USB, just not Main_MiSTer. Could one expand on this and make the SD card a "dumb" bootloader, that simply loads everything from USB? This would also help MiSTer users limit the strain on the SD card slot (no more taking it out just to update Main_Mister), if they got a USB drive. Not sure if this would be helpful for dual-boot though.

Sorry for the off-topic diversion, but I never take the SD out for updating the MiSTer executable. I do it while running: just rename MiSTer to someting else, e.g. MiSTer_old, then copy the new one as MiSTer. (Of course I have to reboot for the change to take effect.)

softtest9 wrote: ↑Wed Dec 01, 2021 10:52 pm 5-600 000 logic elements, did I get that right? With an FPGA like that, I would not be surprised if we got Dreamcast, Nintendo DS, GP32 and a PC core with Pentium II + 3DFX (maybe even Windows XP would be usable). All of those systems are pretty annoying to mess with today, especially bulky old PCs that require IDE drives and so on. So yeah, I will see how things play out, but the price isn't going to stop me if we start seeing these kinds of cores.

It has 256,000 logic cells which translates to roughly 300k~ish of Intel's logic elements. one of the biggest advantages is the much larger BRAM and the onboard SRAM. N64 would likely be possible on it. 3DS would also be possible on it most likely. However, Dreamcast would probably require using HLE due to bandwidth limitations, GP32 is possible but it's not a very advanced system, it's not space concerns stopping GP32 on the MiSTer, it's bandwidth. Pentium II with a 3dfx seems very difficult. Ao486 already is the largest mister core, barely fits, and that's mostly due to sound. Would you have optional awe32/sb16/sb8/adlib?

First of all, the GPU is almost never the bottleneck with emulation. PS2 and beyond are problematic because of high CPU requirements. None of these modern emulators can get full speed without a dynamic recompiler. So that's actually another area where an FPGA might really beat software emulation. All of your other arguments about the PS2 can also apply to the PS1, Saturn and N64. You are of course free to use what you want, but personally I'm more interested in accuracy and low latency, than any kind of glitchy enhancements. That's why I got the MiSTer in the first place.

This isn't accurate. A hypothetical PS2 core would require HLE graphics rendering using the Mali, cpu-bazed emulation on the ARM, and eliminating the need to dynamic recompiling by putting the 128-bit floating point operations on the FPGA. The reason why these separate moving parts with added latency won't hurt the emulation of such a thing is the same reason why it's not worth it to go full FPGA for PS2 emulation. The PS2 and other later gen consoles didn't have such tight timings that a computer would struggle to be cycle accurate. The problem with the PS2 is it's insane over-complexity.

The arguments aren't the same for ps1, N64, Saturn. The previous generation to the PS2 were still very timing-critical.

aberu wrote: ↑Thu Dec 02, 2021 12:39 pm It has 256,000 logic cells which translates to roughly 300k~ish of Intel's logic elements. one of the biggest advantages is the much larger BRAM and the onboard SRAM. N64 would likely be possible on it. 3DS would also be possible on it most likely. However, Dreamcast would probably require using HLE due to bandwidth limitations, GP32 is possible but it's not a very advanced system, it's not space concerns stopping GP32 on the MiSTer, it's bandwidth. Pentium II with a 3dfx seems very difficult. Ao486 already is the largest mister core, barely fits, and that's mostly due to sound. Would you have optional awe32/sb16/sb8/adlib?

Well, 300 000 is quite different from 5-600 000 which I read. I agree that Dreamcast or Pentium II + 3DFX would a very tight fit with 300 000 LEs.

aberu wrote: ↑Thu Dec 02, 2021 12:39 pm This isn't accurate. A hypothetical PS2 core would require HLE graphics rendering using the Mali, cpu-bazed emulation on the ARM, and eliminating the need to dynamic recompiling by putting the 128-bit floating point operations on the FPGA. The reason why these separate moving parts with added latency won't hurt the emulation of such a thing is the same reason why it's not worth it to go full FPGA for PS2 emulation. The PS2 and other later gen consoles didn't have such tight timings that a computer would struggle to be cycle accurate. The problem with the PS2 is it's insane over-complexity.

The arguments aren't the same for ps1, N64, Saturn. The previous generation to the PS2 were still very timing-critical.

I'm not sure what you're on about here. The PS2 most certainly cannot be emulated with cycle accuracy on today's PCs. Maybe the PS1, but a cycle accurate PS1 emulator still doesn't exist yet. What Atohmdiy was arguing was basically that accuracy doesn't matter for the PS2 because muh 4K polygon upscaling, ReShade, texture packs etc.

You can do polygon upscaling and all sorts of graphical enhancements with PS1/N64/Saturn emulators as well, but it's never a good experience (not even with the PS2) with all of the weird graphical glitches that happen. Hybrid emulation using the Mali for HLE graphics would likely not be a good experience either, not so much because of speed, but again because of graphical glitches. Unless you make a Vulkan-based "compute renderer" (basically a software renderer implemented using compute shaders) like the libretro guys did with N64.... maybe.

The time will tell. In my opinion, Mist and mister are only so successful because the price point is right for many.
Replay1 was quite cheap too.
And the more successful a platform is, the more people you will find who will also develop cores for it.

softtest9 wrote: ↑Wed Dec 01, 2021 10:52 pm

aberu wrote: ↑Wed Dec 01, 2021 4:55 pm All in all, if the pricing situation doesn't change for the Kria anytime soon I expect his price point to be minimum $600-800 range. If it is a premium system, that would be awesome. I'm just trying to help people not go out and buy something too fast, definitely take your time as a consumer and do your research when things get announced.

5-600 000 logic elements, did I get that right? With an FPGA like that, I would not be surprised if we got Dreamcast, Nintendo DS, GP32 and a PC core with Pentium II + 3DFX (maybe even Windows XP would be usable). All of those systems are pretty annoying to mess with today, especially bulky old PCs that require IDE drives and so on. So yeah, I will see how things play out, but the price isn't going to stop me if we start seeing these kinds of cores.

Atohmdiy wrote: ↑Wed Dec 01, 2021 7:17 pm Last thing in the "what's the point" argument, with more advanced system you enter the real of 3d, and software emulator can do a lot of amazing thing to increase the graphics. Not sure what this mali gpu could do, but i am sure it's not what my 6900 xt can do If i want to play a 3d ps2 game, i run it with pcx2 with my oled tv with all the fancy option. If one day a ps2 core exist for this new platform, i am pretty sure the mali gpu couldn't do what pcsx2 is doing, so 99% of people would just use pcsx2 with their modern display. I am with the 1% left that would enjoy playing with my crt, and surely this pateform would give the same benefit of the mister : easy setup with a crt (retroarch is giving me headache with my crt...), everything cleaner and more embedded with less lag etc...

First of all, the GPU is almost never the bottleneck with emulation. PS2 and beyond are problematic because of high CPU requirements. None of these modern emulators can get full speed without a dynamic recompiler. So that's actually another area where an FPGA might really beat software emulation. All of your other arguments about the PS2 can also apply to the PS1, Saturn and N64. You are of course free to use what you want, but personally I'm more interested in accuracy and low latency, than any kind of glitchy enhancements. That's why I got the MiSTer in the first place.

I didn't played much with fancy upscalling in modern emulator, but with the few i tried i don't remember seeing much glitch. At least not enough for the average people to being annoyed by it, especially compare to the benefits these filters offer.
I suppose it always be the debate between people that like upscalling in their modern setup vs people with crt that want the "real thing". To my opinion upscalling suck for old 2d games, and for these games nothing will ever replace a crt. For 3d i felt impressed by what these fancy emulator are doing when i saw it, but i will still prefer playing them with my crt, because i like the way crt are rendering low res graphics, even in 3d, and it's more true to the nostalgia.

After don't get me wrong, i am the kind of enthusiast that would buy and support a more powerful platform, even if i have to wait a long time to get something new. I also do prefer low latency, more accuracy and better integration than software emulator. I am just skeptical about the viability of such project right now, especially until the limits of the mister hasn't been reached. I don't see how it could work If you ask the community to buy a more expensive system without something "new" at launched that justify the cost. And saying "it's powerfull enough to emulate X system, we will work on it" to make people waiting years before something is release. I am sure many will get bored before the end... I mean i don't think we will have a n64 or dreamcast anytime soon, i am no fpga developer but it will surely require years of work and dedication, or maybe i am exaggerating ?

That's the whole point. See C65, I would be very surprised if you see cores here that "later" could not be on the Mister.

The price / performance ratio has to be right, otherwise it won't be for the general public. If the number of users is small, one automatically draws much less developer attention.

@mikeJ

Does this mean will see more cores ports to DE10 fork while we wait for the Replay2

The next release of Laxers3As PSX Core seems close and it will have audio enabled:

https://www.youtube.com/watch?v=OPGp4bw2Jws

softtest9 wrote: ↑Thu Dec 02, 2021 2:34 pm Well, 300 000 is quite different from 5-600 000 which I read. I agree that Dreamcast or Pentium II + 3DFX would a very tight fit with 300 000 LEs.

Personally the problem with the dreamcast wouldn't be so much the fit, it would be the bandwidth, the frequencies the FPGA can handle. Xilinx are known to also be faster pound for pound when compared to their Intel Altera equivalents, so maybe this wouldn't be a concern. You can work around this with more multiplexing, but it dramatically increases the space taken and can make it more difficult to achieve proper timings (which to be fair, later gen systems are not as sensitive to).

softtest9 wrote: ↑Thu Dec 02, 2021 2:34 pm The PS2 most certainly cannot be emulated with cycle accuracy on today's PCs.

It could, just not running at full speed, and no emulator does it currently. This is mainly not feasible on modern PC's due to the 128-bit floating point calculations that the PS2 did. You could go cycle accurate, but it may run at like 1FPS or less. It would be a technical simulation, but not a good user experience in the slightest. You wouldn't be able to do it in a way that runs at the same framerate the original hardware did though.

What I meant with the FPGA-based coprocessor is something that RTL Engineering has already proposed before and had started work on, but it didn't have much market viability since those cards are expensive as hell.

I think you misunderstood the intent of what I said, which is my fault.

MikeJ wrote: ↑Wed Dec 01, 2021 11:51 am Can the de10-nano boot from usb?

Far as I know, no. You will need an SD card to boot.