MiSTer Nuked-MD Mega Drive Core

[Rahzadan]

Sorg's release notes state:

"This is a replica of real hardware (unlike functionally similar Genesis core)"

I thought the very definition of replicating a console in FPGA was that the actual logic and chips on the motherboard of the console are replicated accurately? Is the original Genesis core not a functional replica of a Genesis?

Are the other cores not replicas of real hardware? I guess one example comes to mind: On the SNES core, Rendering Ranger R2 requires a ROM patch to work, but on original hardware, it doesn't. If the SNES core were a "replica of real hardware", I assume it WOULD work unpatched since it does on real hardware?

Just trying to understand the difference. Played a few games on this new core and it feels identical to the existing Genesis core!

EDIT: Also to note, if you go to the Nuked-MD GitHub project page, it says "Cycle accurate Mega Drive core". I thought the other cores (well, most of them) were also cycle accurate? (Isn't that kinda the point of an FPGA implementation?)

[rhester72]

Outside of NeoGeo, none of the other cores are replicas of original hardware at a transistor level. They replicate functionality. Cycle accuracy, just like any other form of emulation, is dependent on the relative accuracy of the implementation. FPGA isn't magic pixie dust, it's another way of solving the same problem (with primary advantages in massive scale parallelism like the Amiga chipset, not necessarily accuracy).

[Bas]

There are many ways to write up something that works sufficiently equivalent to a Genesis console to pass as the real thing. It's only by following the actual real hardware for the full 100% that you get an exact replica that's bug-for-bug identical.

The SNES core is a pretty good SNES replacement but not 100% identical to Nintendo's design. Nuked-MD is much closer to Sega's design than the previous Genesis core.

FPGA's aren't magic. As a developer you still have to do a lot of work to replicate existing hardware faithfully. Even then there are compromised to be made. The parallellism that an FPGA enables in contrast to the serial way software emulation works enables closer replicas with less effort but they're not a given.

It's a bit like translating ancient Greek to English. You can get very close but you'll never get it 100% identical.

[mineral55]

I'd love to see a side-by-side comparison of the sound with this new core compared to the original one and other implementations of megadrive/genesis. Judging from my untrained ear I think I can notice clearer sounding FM synthesis in the few games I've tested and the output is much higher as well. I guess it'll get run by MDfourier and we'll have some interesting conversation on all this happening soon. I'm excited!

[Armakuni]

The older Genesis core is based on observed behaviour rather than the actual hardware, similar to software emulators

I5MeDoD.jpeg (31.85 KiB) Viewed 26876 times

[LamerDeluxe]

Original schematics of chips are often not available. One option is to open up a chip (which is a tricky and specialized job), then photograph it using a microscope and painstakingly translate it to the FPGA by hand. I imagine the last part could theoretically be sped up by using AI at some point in the future.

One problem with this method is that an FPGA is not the same as the original chip, this will lead to timing problems. This is what Sorgelig has put a lot of effort into fixing with the Neo Geo core recently and IIRC with this core as well.

The advantage is that your FPGA version will be really close to the original hardware.

Or, as mentioned, you could measure hardware signals with a scope, study or create a software emulation and then adjust the behavior of your FPGA version in response to problems happening with the software being run on it. This can get really close to the original hardware, but will probably have a different electronic structure. I imagine adding enhancements, that the original hardware doesn't have, is easier with this method.

Also as already noted, the main advantage of the MiSTer is its very low latency, as the simulation will run parallel, like the original hardware. Another big advantage is the low power usage that the MiSTer needs, to achieve very high quality emulation of a system.

Personally, I find it fascinating that whole systems (sometimes rare and expensive) are running in hardware on my tiny MiSTer setup.

[Stinky]

Could someone link the old RBF? The update nuked it and it's gone from the repos, I just want to do some A/B.

[LamerDeluxe]

Could someone link the old RBF? The update nuked it and it's gone from the repos, I just want to do some A/B.

Huh, this core replaced the old one already? That is odd. I wonder if it has all of the original options already.

[zakk4223]

The Genesis RBF is still in the distribution repo. You sure you don't have a names.txt that is causing 'Genesis' to show up as "Mega Drive"?

[Stinky]

The Genesis RBF is still in the distribution repo. You sure you don't have a names.txt that is causing 'Genesis' to show up as "Mega Drive"?

Hah yes, of course it was a layer8! Thanks!

[pbsk8]

now we just have to wait for paprium rom

[bazza_12]

I was just about to ask if the Nuked-MD core will now replace the Genesis core, is there any point in the Genesis core being available?

[ryanghappy]

Hey, I'm fascinated by this discussion of observed behavior vs circuit-level sameness.

Does Nuked-MD take up less FPGA space or is it roughly taking up the same amount. I ask this because I know that the other genesis core needed to be split because there's no way to fit Sega CD and 32x all together on that core. Any chance of the nuked MD fitting everything eventually in? Do people actually want that, or is splitting it up still preferred?

[softtest9]

Why should it not be available? You don't have to use it.

[MrSniffles]

now we just have to wait for paprium rom

The comments for that on Kickstarter are... not good

[bazza_12]

Why should it not be available? You don't have to use it.

i wasn't starting an argument, i was just wondering if this is the best, most cycle accurate then is there a point to having the genesis core?

[rhester72]

i wasn't starting an argument, i was just wondering if this is the best, most cycle accurate then is there a point to having the genesis core?

The 'best' argument gets me every time. What's better, red or green? Answer: It depends.

In this case, maybe someone doesn't want CRAM interference, for example - they should choose the older core. Not everyone considers strict adherence to accuracy 'best'.

You'll see this argument a lot more if there's ever a transistor-accurate SNES core. Enhancements like MSU-1 will be extraordinarily unlikely on such a core, and debates will rage. The great news is that you don't have to choose - both cores are available.

[mineral55]

Not everyone considers strict adherence to accuracy 'best'.

Isn't it the better baseline if the point of MiSTer is to recreate these old platforms? I'm always worried that a less accurate core may run games incorrectly that relied on hardware bugs (although then there is the problem of what one might consider the canonical chipset/revision of what's being recreated) but I do think there's an argument to be made for MiSTer to eventually make Nuked-MD-FPGA its primary Genesis/MegaDrive core and anyone who cares more about the featureset of the original core can select an option in update-all to include it if they want.

[bootsector]

MiSTer SNES core is also cycle accurate

[rhester72]

MiSTer SNES core is also cycle accurate

There's a HUGE difference between cycle counts and implementation/integration accuracy.

The Amiga core is cycle accurate in 68000 mode, yet still has flaws in the custom chipset implementation.

I'm not even sure why people continue calling out cycle accuracy in the Nuked core...pretty sure the older core was also "cycle-accurate". That means something VERY specific, and is actually wholly unrelated to emulation accuracy.

[bootsector]

MiSTer SNES core is also cycle accurate

There's a HUGE difference between cycle counts and implementation/integration accuracy.

The Amiga core is cycle accurate in 68000 mode, yet still has flaws in the custom chipset implementation.

I'm not even sure why people continue calling out cycle accuracy in the Nuked core...pretty sure the older core was also "cycle-accurate". That means something VERY specific, and is actually wholly unrelated to emulation accuracy.

Please open a github issue and ask srg320 to correct the README.md of the SNES core then!

[rhester72]

I don't need to. There's already an issue open for Rendering Ranger: R2, which very nicely makes my point.

[mario64]

I was just about to ask if the Nuked-MD core will now replace the Genesis core, is there any point in the Genesis core being available?

Hopefully not. Nuked-MD is missing some features that the Genesis core has such as Adaptive Composite Blend.

[bootsector]

I don't need to. There's already an issue open for Rendering Ranger: R2, which very nicely makes my point.

The game that have issues on real hardware as well? That doesn’t prove any point.

[rhester72]

Sorry, I just read the README after the fact...and I'm confused what's confusing you. It uses a cycle-accurate CPU model, and it's a replica. Just like a lot of cores. Hell, I'm pretty sure the legacy Genesis core also uses a cycle-accurate 68000 model. Perhaps you're confused about what 'replica' means? It's NOT based on a transistor-level decap (and neither is any other SNES implementation ever, up to this point...not even bsnes/Higan).

[pcfreak324]

What is even more confusing is that we now have three Master System cores with different features and compatibility, that all use different rom folders.

• ColecoVision core: best for older games; plays SMS games that use legacy video mode, SG-1000 games and SC-3000 cartridges; has SK-1100 keyboard support

• SMS core: best for newer games; plays SMS and GG games, some SG-1000 games and some SC-3000 cartridges; has support for FM sound, Lightgun, Paddle controls and Multitap as well as an option to increase the sprite limit

• Nuked-MD core: plays some SMS games with video mode 4 and some GG games running in Master System compatibility mode; has support for FM sound; might be the most accurate but should have the same incompatibilities and limitations as a real Mega Drive

[softtest9]

The SNES core is based on hardware documentation and observed behavior. Observed behavior often means using test roms as well as games and comparing results with real hardware.

Nuked-MD, like the NeoGeo core, is based on decapping the actual chips, taking die photos and then translating that to FPGA code. Both methods can certainly lead to a very high degree of accuracy (even cycle accuracy), but the latter is closer to a complete replica. The topic is nuanced further by the fact that you can make mistakes when translating die photos into FPGA code, hence there are still bugs in the NeoGeo core.

You'll see this argument a lot more if there's ever a transistor-accurate SNES core. Enhancements like MSU-1 will be extraordinarily unlikely on such a core, and debates will rage. The great news is that you don't have to choose - both cores are available.

It is entirely possible to mix and match components. If we look at real hardware, there is the SD2SNES/FXPak Pro cartridge with MSU-1 support. There's another cartridge for the MegaDrive that has MegaCD support.

[capitaineflam25]

What is even more confusing is that we now have three Master System cores with different features and compatibility, that all use different rom folders.

• ColecoVision core: best for older games; plays SMS games that use legacy video mode, SG-1000 games and SC-3000 cartridges; has SK-1100 keyboard support

• SMS core: best for newer games; plays SMS and GG games, some SG-1000 games and some SC-3000 cartridges; has support for FM sound, Lightgun, Paddle controls and Multitap as well as an option to increase the sprite limit

• Nuked-MD core: plays some SMS games with video mode 4 and some GG games running in Master System compatibility mode; has support for FM sound; might be the most accurate but should have the same incompatibilities and limitations as a real Mega Drive

Thanks a lot for this summary.

I appreciate having the choice between different, but it's difficult to choose when you don't know those specifities.

[cursedverses]

Once the core is complete (if not already) is the core going to stay true to the original or are these extra features (eg Adaptive Composite Blend) going to be added on top? I personally love the idea of a true-to-the-original core, warts and all.

Maybe if the old core was made available on a separate repository, there could be a way to choose which core to download via a preference in the downloader script?

[LamerDeluxe]

MiSTer SNES core is also cycle accurate

There's a HUGE difference between cycle counts and implementation/integration accuracy.

The Amiga core is cycle accurate in 68000 mode, yet still has flaws in the custom chipset implementation.

I'm not even sure why people continue calling out cycle accuracy in the Nuked core...pretty sure the older core was also "cycle-accurate". That means something VERY specific, and is actually wholly unrelated to emulation accuracy.

Yeah, sadly the timing of the Amiga core is not accurate yet, even in 68000 mode. I made a ticket for rendering bugs in Hybris a long time ago. That game uses a lot of racing-the-beam tricks, with a lot of sprite multiplexing. So it makes a good timing test.