MiSTer FPGA Forum

FPGAzumSpass wrote: ↑Sat Apr 08, 2023 1:17 pm

Atohmdiy wrote: ↑Tue Apr 04, 2023 4:25 pm

I understand that if Robert only want to do console, it's logical to do the fail console after the good one when nothing left, which is the case now. My bet is on the jaguar, but i hope for the 3DO because as i said there still a few correct things in this console.

...
And I wouldn't agree that only fail consoles are left. There are still some left with millions of sales.

OK, that sounds like a clue. Using VGChartz, I listed all of the consoles with 1+ millions sales. I removed all of the ones that already have a core or a core in development, I removed every consoles after the 5th gen and every handhelds with power above 5th gen consoles (PSP, 3DS, PSV).

Results: 3DO, CDi, N-Gage, NDS or the N64

https://docs.google.com/spreadsheets/d/ ... edit#gid=0

My guess, it's the N-Gage.

the_importer wrote: ↑Tue Apr 11, 2023 4:48 pm

OK, that sounds like a clue. Using VGChartz, I listed all of the consoles with 1+ millions sales. I removed all of the ones that already have a core or a core in development, I removed every consoles after the 5th gen and every handhelds with power above 5th gen consoles (PSP, 3DS, PSV).

Results: 3DO, CDi, N-Gage, NDS or the N64

https://docs.google.com/spreadsheets/d/ ... edit#gid=0

My guess, it's the N-Gage.

NDS is actually more powerful than any 5th gen console including N64.

H6rdc0re wrote: ↑Tue Apr 11, 2023 6:01 pm

NDS is actually more powerful than any 5th gen console including N64.

It's debatable, the N64 does some things better than the NDS while the NDS does somethings better than N64. Mind you, there's a bit of GBA DNA in the NDS, so between the two, my money would be on the NDS.

the_importer wrote: ↑Tue Apr 11, 2023 4:48 pm

FPGAzumSpass wrote: ↑Sat Apr 08, 2023 1:17 pm

Atohmdiy wrote: ↑Tue Apr 04, 2023 4:25 pm

I understand that if Robert only want to do console, it's logical to do the fail console after the good one when nothing left, which is the case now. My bet is on the jaguar, but i hope for the 3DO because as i said there still a few correct things in this console.

...
And I wouldn't agree that only fail consoles are left. There are still some left with millions of sales.

OK, that sounds like a clue. Using VGChartz, I listed all of the consoles with 1+ millions sales. I removed all of the ones that already have a core or a core in development, I removed every consoles after the 5th gen and every handhelds with power above 5th gen consoles (PSP, 3DS, PSV).

Results: 3DO, CDi, N-Gage, NDS or the N64

https://docs.google.com/spreadsheets/d/ ... edit#gid=0

My guess, it's the N-Gage.

The N-Gage would actually be really cool. So would an environment for J2ME games. Old cell phone games are sort of getting forgotten.

NDS is a no go in full FPGA. I think I remember FPGAzumSpass tried but the speed for 3d was super slow or am I remembering wrong? Maybe an Hybrid core build with the help of ChatGPT? Why are you all laughing I can see you there!

If I recall correctly, NDS could possibly work but not be perfectly accurate to the original hardware. What I don't understand is how the touch screen would work. I am guessing the user would need to use a mouse.

I still think it will be a Jaguar core.

in my old DS core you could either use a mouse or the analog stick for the touch screen. At least for the games i tried, one of them always worked good:

• for games that only need it occasionally, analog stick is ok
• for games that need it all the time(with touchpen), mouse works good
  PS5 pad with touchpad could also be fine, but i never tried that.

The core i made back in 2020 suffered from several issues:

• it was not a fully cycle accurate design
• the CPU was still 3 stage from GBA, not 5 stage as it should be, leading to not being able to reach the required clock speed
• the DS has many parallel internal RAMs with insane total bandwidth and very low latency. DDR3 + dual sdram on MiSTer could cover the bandwidth probably, but never the timing
• The full design is very large, due to it being more or less 2 times GBA(2xCPU, 2x2D-GPU, 2xDMA, ...) + 3D rendering and geometry, so might also be hard to fit logic wise

That being said, it might be possible to make it run about as good as GBA or PSX, but not fully 100% cycle accurate due to the RAM. And very likely savestates would not fit in, making the debugging a nightmare.

FPGAzumSpass wrote: ↑Wed Apr 12, 2023 4:51 am

in my old DS core you could either use a mouse or the analog stick for the touch screen. At least for the games i tried, one of them always worked good:

• for games that only need it occasionally, analog stick is ok
• for games that need it all the time(with touchpen), mouse works good
  PS5 pad with touchpad could also be fine, but i never tried that.

The core i made back in 2020 suffered from several issues:

• it was not a fully cycle accurate design
• the CPU was still 3 stage from GBA, not 5 stage as it should be, leading to not being able to reach the required clock speed
• the DS has many parallel internal RAMs with insane total bandwidth and very low latency. DDR3 + dual sdram on MiSTer could cover the bandwidth probably, but never the timing
• The full design is very large, due to it being more or less 2 times GBA(2xCPU, 2x2D-GPU, 2xDMA, ...) + 3D rendering and geometry, so might also be hard to fit logic wise

That being said, it might be possible to make it run about as good as GBA or PSX, but not fully 100% cycle accurate due to the RAM. And very likely savestates would not fit in, making the debugging a nightmare.

Couldn't you run Arm code on the Arm core? I read in a white paper Arm v7 is backwardscompatible with Arm v6. Perhaps that also could work for 3DO, GBA and NGPC.

Data exchange from ARM Core to the FPGA fabric is too slow to have that as accurate solution.
Also it only solves like the 3rd biggest problem of a DS core, but introduce several new.

FPGAzumSpass wrote: ↑Wed Apr 12, 2023 3:28 pm

Data exchange from ARM Core to the FPGA fabric is too slow to have that as accurate solution.
Also it only solves like the 3rd biggest problem of a DS core, but introduce several new.

Guess the DS is out of the question then. That being said, what about the N64? As the first one to successfully bring us the core of a fully 3D system (PS1), in your opinion, what makes the N64 out of reach compared to the PS1, Saturn and 32X?

As an aside, not all the pre-5th gen systems not on MiSTer yet are "failures". For example, before the Famicom was released the Epoch Cassette Vision dominated the Japanese market. That one only recently had it's games dumped (a difficult process due to chips being on the carts and not the console) that as yet nobody has emulated.

There is also the successor, Super Cassette Vision that wasn't as popular but did still have some fun looking exclusives on there of licensed properties such as Doraemon, Lupin III and Dragon Ball. Ironically that one would be much easier to make a core for.

They may not be the big, exciting, glamourous systems but there are definitely some older consoles we don't yet have that aren't without merit and have some fun, albeit often basic, games in their libraries.

the_importer wrote: ↑Wed Apr 12, 2023 3:50 pm

Guess the DS is out of the question then. That being said, what about the N64? As the first one to successfully bring us the core of a fully 3D system (PS1), in your opinion, what makes the N64 out of reach compared to the PS1, Saturn and 32X?

There is only one real issue that needs to be worked against: the CPU at 93Mhz, which is nearly 3 times as much as the PSX clock rate and it's 64bit and it has a FPU.
It's very unlikely that such a CPU can work in the DE10-Nano, at least 100% accurate with timing closure.

The other parts would very likely work out well, including the RDRAM memory which is often said to be super fast, but it's not. The DDR3 on the DE10-Nano has higher bandwidth and comparable latency.
Again, no 100% accuracy here, due to the nature of the DDR3 being shared with the ARM and scaler making it somewhat random, but close enough to make games work i guess.

Overall space could also be an issue. The N64 doesn't have many components, mainly CPU with FPU, second CPU with Vector coprocessor and GPU. The other parts are relativly small.
But each of these main components is much heavier than their counterparts in PSX.

Another big hurdle is the time consumption that it takes.
We already see how difficult it is to develop a 5th gen multi cpu system with e.g. Saturn and Jaguar.
Building such a system with about 3 times higher clock speed and wider datapath (more logic) is not easy.

So are you having fun watching us all scratch our heads wondering what the core is going to be @FPGAzumspass?

FPGAzumSpass wrote: ↑Wed Apr 12, 2023 5:33 pm

There is only one real issue that needs to be worked against: the CPU at 93Mhz, which is nearly 3 times as much as the PSX clock rate and it's 64bit and it has a FPU.
It's very unlikely that such a CPU can work in the DE10-Nano, at least 100% accurate with timing closure.

The other parts would very likely work out well, including the RDRAM memory which is often said to be super fast, but it's not. The DDR3 on the DE10-Nano has higher bandwidth and comparable latency.
Again, no 100% accuracy here, due to the nature of the DDR3 being shared with the ARM and scaler making it somewhat random, but close enough to make games work i guess.

Overall space could also be an issue. The N64 doesn't have many components, mainly CPU with FPU, second CPU with Vector coprocessor and GPU. The other parts are relativly small.
But each of these main components is much heavier than their counterparts in PSX.

Another big hurdle is the time consumption that it takes.
We already see how difficult it is to develop a 5th gen multi cpu system with e.g. Saturn and Jaguar.
Building such a system with about 3 times higher clock speed and wider datapath (more logic) is not easy.

Damn, makes me wonder how great the N64 could have been if it had been a CD-ROM system, the PlayStation brand could have died in that same generation.

Well, I guess this leaves us with 3 possible candidates then

Is a computer core a possibility? I would personally love a more polished X68000 core, but I guess that's probably not a million-seller (can't find sales figures even in Japanese but one blog said 350,000-450,000 at the most).

PC98 looks like it might have sold more than a million units though.

Either one would earn an instant donation for me.

Still waiting patiently for the Apple IIgs.

the_importer wrote: ↑Wed Apr 12, 2023 6:41 pm

Damn, makes me wonder how great the N64 could have been if it had been a CD-ROM system, the PlayStation brand could have died in that same generation.

Well, I guess this leaves us with 3 possible candidates then

Biggest problems of the N64 are it's very limited texture cache and super slow unified memory architecture. Nintendo being Nintendo they went with 4MB RDRAM at the last minute when they should have gone with 8MB RDRAM from the start. Also using dedicated RAM and VRAM so 4MB each and a bigger texture cache would have brought the N64 closer to Dreamcast than it would have been to PS1. Ultimately with the final product they wasted so much promise and were not that far beyond PS1. Both it's CPU and GPU were way beyond PS1 but were dragged way down. A pattern they never stopped with as each console before and after were seriously hamstrung by Nintendo in some way.

SNES originally had a 68000 @12MHz but because Sega used a 68000 for the Mega Drive/Genesis out of spite they went with the very limited Ricoh 5A22 instead. They downclocked the Gamecube Flipper GPU from 243MHz to 162MHz and halved it's RAM from 48MB 1T RAM to 24MB. Wii would have had motion plus like technology from the get go but they opted to take it out and sell it as an add-on later. Wii U originally would get a GPU comparable in performance to the Radeon HD4870 but it got a very limited and crippled GPU instead. Switch has it's Tegra X1 downclocked by a lot in order to save on using a bigger and better battery. Sorry just a little rant.

I think we can rule out any of the Japanese computers, since the developer doesn't speak Japanese. I could however see a deeper rewrite of ao486 happen, or a replacement.

H6rdc0re wrote: ↑Wed Apr 12, 2023 8:20 pm

SNES originally had a 68000 @12MHz but because Sega used a 68000 for the Mega Drive/Genesis out of spite they went with the very limited Ricoh 5A22 instead.

The SNES basically inherited its CPU from the NES. The SNES was originally meant to be backwards compatible with NES games.

the_importer wrote: ↑Tue Apr 11, 2023 4:48 pm

Results: 3DO, CDi, N-Gage, NDS or the N64

I think it's the 3DO, it sold a few millions and has some unique quirky games that sold well.

But even if I'm wrong. I'm happy to have a new core coming up for any of those systems. Can't wait to dig into a new game library!

It’s the N64.

If anyone was going to take on the N64, it had to be Robert! I just read his Patreon post and setting my expectations accordingly. This is going to be a journey!

I think most people would be happy with an N64 core that plays well even if we can't have one that's fully accurate on the DE-10 Nano's specs.

Very exciting news indeed!

There’s no way… I can’t believe that he’s actually attempting it! The absolute mad man!!

Even if it ends up not fitting onto the DE10 Nano, it would still be an unbelievable accomplishment!

Godspeed Robert!

His update sounds very, very optimistic.

Sounds like he has some ideas on how to work around certain previously thought to be hard limitations.

Even thinks he'll get a playable homebrew version sometime this summer.

Can't wait to see how this unfolds.

Awesome news.

Woohoo! This is great, looking forward to it!

support his patreon https://www.patreon.com/FPGAzumSpass/

It’s N64. Now this thread can be deleted.

Wow! I told you it was the N64! Oh wait, let me go back and edit all my Jaguar prediction posts first... If anyone can manage to pull this off, it's Robert Peip. Amazing news, even if it ends up not working out, which for the first time, I have some kind of confidence that it might!

Now that the N64 is in the realm of possibility, I am more convinced that a faster x86 core with a Voodoo card is possible. The original N64 emulator UltraHLE ran on a modest PC with a 3DFX card. It did this by translating the N64 3D calls to the extremely similar Glide 3D calls and it just worked almost perfectly. So I would say that once the N64 is done, some of the 3D portion might be able to be re-purposed in a Voodoo PC card. It can probably all fit with a new clean and lean design. The beauty of this would be the ability to change it later to add performance enhancements to both the CPU and GPU since PC software is not exact hardware specific, other than the required drivers for the hardware. Maybe someday, we can play UltraHLE on our PC core in the MiSTer. Just kidding - UltraHLE requires a P2-233. I guess before trying to make a Voodoo capable core, it would make sense to look at the PC requirements for the games. One of the earliest games was for DOS called Whiplash. It was hard to find, but I think this game will play on a 486 DX2/66. GLQuake requires a Pentium 90 - maybe just barely possible in comparable performance with a custom MiSTer CPU. Same with Moto Racer. Same with Need For Speed 2 SE. It seems like a Pentium 90 was the main computer when the Voodoo came out.