N64 Core Possibility? (Update! Robert Is Working On This Core!)

[pgimeno]

Regarding hybrid emulation, even if it did work, imo it would kind of defeat the purpose of FPGA. The goal for FPGA imo is to 100% replicate the hardware in HDL to preserve for future generations.

I think you won't need more than one hand to count the cores for which this is true. The AY-3-8500 core is one of the few that do that. The Gigatron core is another. Can't think of any other right away. Every core that uses a Z80, for example, is using the T80 module which is a recreation not based on the original hardware.

[Hodor]

[...] Every core that uses a Z80, for example, is using the T80 module which is a recreation not based on the original hardware.

Out of curiosity, why is that? Is not the actual Zilog Z80 currently decapped and traced?

[pgimeno]

Out of curiosity, why is that? Is not the actual Zilog Z80 currently decapped and traced?

I'd like to know as well. Maybe T80 was around before the work on Z80 was completed, and no one felt like the effort of creating one based on the decap was worthwhile, having T80 already. Or maybe T80 is more FPGA-friendly, like using fewer resources or something like that. I can only conjecture.

[KremlingKuthroat19]

There is about zero chance a board at the DE-10's price point will support N64 in decades. The clocks just aren't there.

Decades is an absurdly hyperbolic take. The FPGA industry is growing by leaps and bounds because of machine learning. To think that Moore’s Law applies to everything but FPGAs is beyond silly.

Robert Piep mentioned it may be just a few years away on his interview with Arduino earlier this year and he certainly knows more than me. Btw, the MiSTer is over 5 years old now. I have a hard time believing FPGA technology will be the same 5-10 years from now.

Hopefully that’ll change, but even if the performance and price is feasible, developer productivity doesn’t scale with Moore’s law, while the number of work hours required to implement the advanced systems a new platform allows probably do.

That's a great point. I completely agree that advanced systems will be much harder to implement not necessarily due to constraints on FPGA hardware, but due to manhours. However, we do have an almost complete N64 core. It took the developer 4 years, but it's being close to accomplished by one man. Cycle accurate 5th/6th gen consoles is a very tantalizing prospect. I can see teams being formed around this in the next 5-10 years to tackle huge projects like a Dreamcast or a PS2 one day. I'm not claiming that this happen in the near future, but just saying it won't be in the distant future either imo.

Regardless, I think FPGA gaming's future is so bright that it's frankly blinding. Ignoring the million dollar question of getting a new board to run more advanced consoles, we'll get a Saturn core in the next year or 2 and who knows after that. By decade's end I suspect that most everything that can run on MiSTer will run on MiSTer. We've seen a big influx of developers recently, especially on the arcade side of things and we're getting to the long tail of diminishing returns for consoles and PCs. Being an owner of the MiSTer since summer of 2020, I can say that it's pretty miraculous when you consider how many cores we've gotten in that time. Every year there's huge surprises and the value proposition of the MiSTer grows by leaps and bounds.

[dmckean]

[...] Every core that uses a Z80, for example, is using the T80 module which is a recreation not based on the original hardware.

Out of curiosity, why is that? Is not the actual Zilog Z80 currently decapped and traced?

A more vertical oriented implementation is going to consume less resources in the core with the trade off being that it needs to run at a higher clock speed. As long as it executes all the instructions in the same manner and with the same timings as the original it doesn't matter.

[Newsdee]

Regarding hybrid emulation, even if it did work, imo it would kind of defeat the purpose of FPGA. The goal for FPGA imo is to 100% replicate the hardware in HDL to preserve for future generations. A hybrid approach opposes this.

That's debatable. Yes keeping a core as 100% HDL will help port it to other boards, but there is nothing wrong with taking a hybrid approach as long as it helps push beyond the limits of the DE10.

That said, from what I have seen so far it makes the design more complex without a lot of extra benefit. I wonder whether daisy chaining 2 or 3 DE10s via GPIO might be a better solution than using the embedded ARM.

The main limit is bandwidth between components, so that still wouldnt work for N64, but it may allow designs with more LEs if eg the scaler is moved to a 2nd board.

[seastalker]

... you mean a MiSTer MiSTer?

I actually like the idea. I'm not qualified to suggest it would work or not, but if people stack pi's in tandem for crypto, maybe a double-mister can do more advanced cores. I don't envy the developer to take on the task.

[rhester72]

Been suggested and shot down many times on this very forum - not enough external data lines to do it (and what's there isn't nearly fast enough).

[thorr]

There is actually a much simpler and cheaper solution. I already solved it and am already enjoying my fully developed N64 core. It works 100% correctly too. Get a used N64.

[Mr. Encyclopedia]

Used hardware is well and good, and these days with everdrives and ODEs and recapping we can extend the life of old hardware further than previously hoped, but nothing lasts forever.

[TobyRieper]

Exactly. This is the reason im getting a mister again...sold mine in 2020 when i needed funds.

[eldritch]

In case anyone missed it, Robert's secret core is the N64! Of course expectations still need to be in check but he's gonna give it a go! Even if some workarounds need to be in place to make it happen it'll be cool to have it.

https://twitter.com/AzumFpg?ref_src=tws ... de%3Dfalse

https://www.patreon.com/FPGAzumSpass

[Waifu4Life]

Well that topic aged poorly.

[Neocaron]

It did and I'm really happy about it. I do expect some hack ao486 and PSX style for the core to get it to full N64 speed though, but I don't mind! As long as the visible in game accuracy is 99% there. I think it is fair to say that after the PSX core, we can trust him to deliver.

[dcubed]

Well that topic aged poorly.

Bit early to say that yet, even Robert himself said that he's not sure if it's possible on the current DE-10 nano.

But either way, the foundational work here will mean that an N64 core will basically be ready to go as soon as we get a viable successor to the DE-10 nano, so it's still very exciting either way.

[rhester72]

Well that topic aged poorly.

Bit early to say that yet, even Robert himself said that he's not sure if it's possible on the current DE-10 nano.

But either way, the foundational work here will mean that an N64 core will basically be ready to go as soon as we get a viable successor to the DE-10 nano, so it's still very exciting either way.

To be honest I think the bigger thing that may come of it is an actually viable and accurate software emulator which has never before existed.

[mcf81]

To be honest I think the bigger thing that may come of it is an actually viable and accurate software emulator which has never before existed.

Accurate? Yes, definitely ... Robert's emulator may be the most accurate N64 emulator ever created, the "bsnes of N64". But that is also applied to the performance (bsnes is brilliant but not exactly super performant) -- I don't believe Robert intends for it to be usable for anything other than testing. His latest (amazing!) screenshot of his emulator running Ocarina of Time also shows it was running at ~ 30% speed. I doubt he would be interested in refactoring it to run at a normal speed since that is not at all what it was developed for, but perhaps someone could come behind, fork the project, and optimize it for use outside of testing.

Also worth reiterating is that even if Robert hits a wall with N64 on DE-10 Nano, the community will have a tremendous incentive to find the next platform (playing N64 and possibly even DS with all the work Robert has also put in there!). Although off-the-shelf FPGA boards are out of the question due to the expense, perhaps someone in the community could do a for-profit crowdfunding of a workable FPGA platform ala Analogue...

[Newsdee]

the community will have a tremendous incentive to find the next platform (playing N64 and possibly even DS with all the work Robert has also put in there!).

This 100%. MiST and later MiSTer gave a home to many cores that came before them. So having cores that don't fit on a. MiSTer will eventually make a bigger device interesting.

perhaps someone in the community could do a for-profit crowdfunding of a workable FPGA platform ala Analogue...

I wouldn't hold my breath on this one. Analogue (an established name) still was unable to make a device surpassing the MiSTer's FPGA for a good price. Why would a crowdfunding be any better?

[mcf81]

I wouldn't hold my breath on this one. Analogue (an established name) still was unable to make a device surpassing the MiSTer's FPGA for a good price. Why would a crowdfunding be any better?

I’m probably just dreaming big but with all the amazing things coming from the community these days it’s an easy thing to do!

[AtomicShroom]

The DE-10 Nano has been around since 2017, 6 years ago. Considering how fast things typically evolve in the field of computers, I find it kind of surprising that Terasic has yet to release a more powerful iteration, and that we're not even hearing about one being in the works. I'm not familiar with how FPGA processors are made, but can someone who does explain why things are this stagnant?

[Nickie]

it's not about power, it's about the size of the chip itself, some can contain more gates than other, so there ARE bigger dev boards, but they are more expensive, you could spend 12k for some of them babies, we would totally port mister to them but who would buy them?

so the MiSTer project might move on to a better FPGA dev board, one day, if necessary, but right now since FPGAs aren't really a mainstream appliance, prices don't go down, in fact, even the de10's price goes up over time, and would be a lot more expensive if Intel wasn't subsidizing it for education

[grizzly]

The DE-10 Nano was and is mostly directed at the beginners to start doing FPGA stuff, aka draw them in so they buy more FPGA´s for their projects and then to their job projects.
That is one reason that so many things is built into it like network connector/hdmi and so on.
At the same time that is one reason why there has not been any larger/faster FPGA´s put on it/or a new similar board has been made.
The DE-10 is simply big and fast enough for beginners and even for many really advanced things!

[Antoine.WG]

It's worth noting that the N64's MIPS R4300i CPU contained about 1.7 million transistors, over 10x the logic elements in DE-10 Nano's FPGA (110,000), and the RSP transistor count is 2.6 million. There are undoubtedly some optimizations that can and have been made but it may never be possible to fully emulate the N64 at full speed on this hardware.

[cacophony]

It's worth noting that the N64's MIPS R4300i CPU contained about 1.7 million transistors, over 10x the logic elements in DE-10 Nano's FPGA (110,000), and the RSP transistor count is 2.6 million. There are undoubtedly some optimizations that can and have been made but it may never be possible to fully emulate the N64 at full speed on this hardware.

Good thing a logic element is significantly more capable than a transistor

[willis936]

imo Xilinx offerings are crushing Altera. The Kria KR260 is a bit expensive at $350, but its XCZU5EV has more than twice the LEs than the Cyclone V. The DE10 Nano wins for having an insane number of user GPIO exposed. I don't know of any other boards that break out 80 pins. There are cheaper Kria options with even less broken out I/O, but I can't imagine anyone would be happy with even less than half the DE10 Nano's GPIO.

Also, it's essential that retooling isn't required to port cores, otherwise stuff will fall through the cracks. Altera is likely the safe choice.