Faster CPU for Minimig core?

[lordoftime79]

I was wondering if anybody is working on the core still and more so the CPU speed? I love the core on the Mister but after comming from using Vampired Amigas find the CPU speed a little dissapointing and would love to see improvements.

[limi]

If you want Vampire-level performance, you'd need to do something like using the ARM CPU to do 68K emulation as part of the core. I know Rok Krajnc (one of the Minimig authors) got a MiSTer and was looking into it around August last year, but his blog hasn't updated since then. (I won't link to it since it's infested with malware right now, probably an unpatched Wordpress install or something)

[LamerDeluxe]

Or 680x0 emulation

[Michael1260]

For me two features will be nice on Minimig :
- a better / faster CPU (I also have a Vampire so I can you use it for "heavy" tasks but ...)
- "native" ethernet support from the Mister (PPP is so sloooowwwwwww)

[throAU]

Once you’re in the realms of non-standard amiga hardware the timing issues to make old software work properly kinda go out the window and you may as well just run UAE?

[Michael1260]

If you have a 68030 CPU, you can probably add a screen on the OSD between 68000, 68020 and 68030

[Bas]

A fully decked-out 68030/40 with FPU, MMU. would open up many doors, not just Amiga related. A lot of early UNIX workstation gear was also based on the bigger 68k's, AMIX may even run, not to mention Apple Macintosh with a bit more oomph than the MacPlus. But beggars can't be choosers so I patiently await what the gurus produce.

[rhester72]

AMIX _requires_ an A3000 with tower ROMs, period, or it will not function. It's also worse than useless current-day. There's barely even a network stack. Play with it on WinUAE and tell me you're still interested more than 10 minutes later.

Couldn't agree more with an earlier poster - if all you want on planet Earth is the fastest Amiga possible, you're looking in the wrong place with MiSTer. That's not what prosumer-level FPGA shines at. Get a decent laptop, run WinUAE, admire your 10,000MHz Amiga.

[kolla]

Not so, AMIX also runs on A2000 with A2620 or A2630 cards, Commodore even shipped these as A2500UX.

https://bigbookofamigahardware.com/bboa ... aspx?id=21

[rhester72]

Hrm...right you are. Never saw one of those in the States. Thanks for that!

[lordoftime79]

AMIX _requires_ an A3000 with tower ROMs, period, or it will not function. It's also worse than useless current-day. There's barely even a network stack. Play with it on WinUAE and tell me you're still interested more than 10 minutes later.

Couldn't agree more with an earlier poster - if all you want on planet Earth is the fastest Amiga possible, you're looking in the wrong place with MiSTer. That's not what prosumer-level FPGA shines at. Get a decent laptop, run WinUAE, admire your 10,000MHz Amiga.

such a waste - it could do so much more but I guess there are too many small minded people like you.

[Bas]

I'm not looking for useful UNIX on a MiSTer. I have FreeBSD 13 for amd64 coming for that this month. I'm also not looking for a fast Amiga. What I am looking for, and what FPGA is amazing at, is faithful reproduction of rare, old and fragile computers so that their workings can be experienced and studied without breaking the bank or requiring training in advanced electronics repair. Reasons similar to why museums opt to include replicas in their collections.

Why is it legitimate to create a cycle-exact version of the A500 but not the A3000? I obviously get the retro gaming appeal, of the 500 but there is so immensely much more to the history of computing than just games. Amiga was an amazing platform and all of it deserves to be conserved as faithfully as possible. Original hardware will inevitably disappear beyond the reach of regular hobbyists, so FPGA reimplementation has a niche to fill here.

[rhester72]

@Bas Mostly because cycle-exact mattered on the original Amiga design due to the incredibly close coupling of the CPU bus with the custom chipset. Once the CPU clock was decoupled from the system bus (pretty much every accelerator ever except the A1200), it no longer mattered, which is why so many early games and demos require cycle-exact to function properly but applications and pretty much anything written after 1990 does not.

How does one create a 'cycle-exact' A3000, anyway? For which CPU? For which ZIP type? Do we account for burstable RAM? Do we use the broken SCSI controller for authenticity, or the fixed one? That's the entire problem with big-box systems, particularly Amigas...there isn't "one", and nobody's going to be satisfied with anything short of the entire kitchen with sink (which is what WinUAE is/has become, and is UTTERLY impractical/impossible on MiSTer/FPGA).

@lordoftime79 Sorry you feel that way - by all means, please illuminate me with the use case where a super-fast decoupled Amiga bus on FPGA will provide more benefit than WinUAE running on, well, anything. I'm dying to hear it.

[kolla]

Fast 68k CPU + ECS/AGA on FPGA is the only useful hardware config missing, and that is easiest done by moving CPU to the ARM using JIT software emulation.

[Bas]

That's quite a bold statement: the only useful hardware config. Being an overclocked 68k adds nothing functionally. A 60030 with MMU and FPU would. WinUAE needs a PC with Windows. That's completely incomparable to the instant-on hassle-free near native feel in a tiny box of the MiSTer.

Which model? Simply the most basic initially released version of the A3000 would do. If that finds uptake, additional features will surely get added over time. It's far too simple to dismiss something because you personally don't see a use for it.

[rhester72]

No, I just understand the complexities of FPGA design and how insanely difficult it is to 'swap' components. I'd argue there's a MUCH stronger use case overall for more C64 peripherals than there is for a fantasy Amiga rig, but to each their own. I wouldn't exactly consider a Windows-based laptop a barrier to entry in 2021, nor would I come anywhere NEAR describing anything about MiSTer 'hassle-free'.

It's never going to happen, but I won't derail the thread further.

[lordoftime79]

so I take it you guys havnt seen a Vampire then? thats my use case, thats not uae and thats fast as fuck. I dont see any reason why Mister cant be fast too this thing you go on about cycle exact isnt an isssue it dosnt stop the Vamped amigas being backwards compatible but makes them able to play alienbreed 3d2 full speed, Quake at a decent speed and lots of other good stuff. you are obviously not an "Amiga" guy so you wouldnt get it @rhester72

[rhester72]

I'm just going to chuckle out loud that I'm not an 'Amiga guy'. I'm sure Google works for you as well as anyone else.

As for Vampire...why didn't you just say you wanted a free one? I'm sure the creators would be delighted to share the source code, just ask! (And for the record, yes, WinUAE can do all of that, and PPC to boot...just sayin'.)

Kind of tiring of this utterly pointless discussion now, so marking this thread perma-read on my end. There's never any end to this craziness.

[lordoftime79]

I have both types of Vampire thank you.

[Bas]

How is a request for an accurate FPGA-simulated Amiga 3000 craziness? Software emulation did Amiga long before MiSTer was even a thing. By your reasoning the whole project should not have happened because there already was a working alternative.

Sure an A3000 may be darned near impossible to do (I don't know) and the value of it may be low in terms of enabling authentic Amiga use cases that weren't already covered by the existing Minimig. That's all fine with me. Like I said, I don't have the skill to do it nor the cash to hire the devs for it so I'm not making any demands here.

What sits badly with me is the condescending tone with which a legitimate question apparently needs to be shot down and any follow-up disqualified and ignored.

If that's how you feel, why respond in the first place?

Heck, I don't really see where @lordoftime79 is coming from either but I understand their request. There is indeed a valid FPGA-based alternative for "bloody fast Amiga-esque" system called Vampire. That obviously leads to the question: can a similar speed bump be done on MiSTer? Well I guess not without rewriting the whole Amiga core, and the same probably goes for real big-box Amigas that did exist.

My reason for interjecting was to propose a middle ground between Minimig and Vampire. I'm not in the market for a Vampire, and a 68030 would be faster than what's available now. If an as-authentic-as-possible facsimile of the A3000 could be built from that, that would be awesome.. at least to me.

And if it couldn't, a 68030 would be a great step in enabling other things besides Amigas like NEXT computer or early SGI gear. So here I am, declaring interest if ever someone would seriously consider looking into trying for big box Amigas. And sure I'd chip in if needed.

[rhester72]

@Bas Several reasons. First, what _is_ an A3000? How much RAM should it be populated with? Do you consider an accelerator slot? Do you want the full/authentic SCSI controller? What about Picasso96? Do we emulate the scandoubler? And that's just _one machine_. 4000, anyone? How about a Video Toaster? You can tap every niche out there - or you can use what *already exists and works perfectly fine* without reinventing the wheel and trying to do the to-a-hammer-every-problem-looks-like-a-nail thing.

The ONLY thing FPGA brings to Amiga is cycle-exact concurrency, which was only actually valuable (from a software compatibility standpoint) on the A500 (and to a lesser degree the A1200). EVERY other machine, with EVERY non-68000 CPU (stock or not), splits the busses and negates the need (or benefit) for close coordination between the CPU and chipset...and thus takes away pretty much every single advantage MiSTer brings to the table except CRT output (and possibly someday use of an actual floppy drive if desired), both of which become largely meaningless as you move towards a 'big box' desire.

The same 'legitimate' questions have been asked time and again, here and on the old forums, right along with the "how about we recreate a DVD player?' and 'why can't ao486 run at a terahertz, IT'S HARDWARE!!!' threads. Right problem, perhaps - but *wrong platform*.

re: Vampire...yeah. An audience of hundreds, literally, solving a problem nobody had. You get one...great. What now? You...run Quake, a title designed from the ground up and best enjoyed on PC, and ported to FAR more capable platforms? Yes, OK - there's the "because I can" aspect of it, true...but anyone with both the technical capability *AND* niche-within-a-niche-within-a-niche interest in it isn't going to create something (like effectively 'porting' Vampire to MiSTer) because someone asked...they'll do it for themselves, and if anyone else wants a play, great.

68030 is no faster intrinsically than an '020...outside of burst mode, FPU and MMU, they are logically _the same chip_. This is why most software emulation guides recommend the lower-complexity '020 unless you REALLY NEED the other features. The '030 was faster back in the day because it was a newer part with more advanced manufacturing techniques that allowed it to be clocked higher...a limit that does not exist in FPGA. I agree it'd be REALLY damned cool to see a NeXTcube or Indigo...I could HELLA get behind that, due to their rarity and exotic nature...but again a) they were more or less point-in-time hardware that didn't get a lot of bus-level expansion devices that complicate things, and b) it'll still take someone with the know-how AND the itch to scratch, which won't come about because of a wishful forum post.

I know I said I was done, so sorry for the noise, but I wanted to clarify that my 'meh' about all this is based in experience and reality, not a desire to piss on anyone's parade.

[limi]

My reason for interjecting was to propose a middle ground between Minimig and Vampire. I'm not in the market for a Vampire, and a 68030 would be faster than what's available now. If an as-authentic-as-possible facsimile of the A3000 could be built from that, that would be awesome.. at least to me.

To bring this back on topic and away from the bickering: The current Minimig core is faster than an Amiga 3000, even without adding an FPGA FPU.

Compared to a 25MHz 68030 Amiga 3000, AIBB reports the MiSTer core as having:
• 2.19× the performance on Integer Math tests
• 3.17× the performance on Graphics Rendering tests
• 2.05× the performance on Floating Point Math tests — note that MiSTer does not have an FPU, unlike the Amiga 3000

Original post

So the only thing “missing” is MMU support, which pretty much no Amiga software made use of, except for some highly specialized cases — e.g. Enforcer for development. There are a very small set of software that refuses to run without an FPU, no doubt — but I haven’t encountered any so far in my testing that does not also include an ’020 compatible binary.

The difference between the 68030 and the 68020 is less than you might think, and MiSTer’s ’020 is massively faster than the 25MHz Amiga 3000.

Also note that the 68030 was almost identical to the 68020 overall, just with an additional 256 byte data cache, a process shrink, and an added burst mode for the caches. So 020 vs 030 as far as MiSTer goes is kind of moot. Any instruction set optimizations did not arrive until the 68040.

Original post

The original question here was if the MiSTer could get to Vampire-like speeds — which is possible, but there’s no available code that attempts to do that yet. But I'm sure it will happen eventually.

[LamerDeluxe]

A 68030 doesn't have an FPU, the 68040 and 68060 do. For the 68030 (and 68020) you can add a 68881 or 68882 (I had the latter one in my Amiga 2000). Those are even capable of 80-bit floating point calculations.

I don't see the usefulness in implementing a SCSI controller when you use VHD images and don't have an external port for it anyway. MiSTer already has a scandoubler, I don't expect a hardware scandoubler to have any influence on any cycle-exactness, it is just a buffer. I used to have one in my Amiga 2000.

A Video Toaster implemented in FPGA would be awesome, but likely way too much work for something to just play around with. Though I'm also all for cores of exotic systems that are impossible to experience otherwise.

680x0 FPU support would be great to be able to properly run the productivity software that used it. Running Sculpt 4D or Imagine on the MiSTer would be great, for instance (I used those a lot) just as being able to run something like Wavefront on an early SGI machine. I can imagine implementing the graphics hardware of an SGI machine would be a massive undertaking though.

I agree that cycle-exact implementations would probably not be noticeable with these faster machines. IMO being able to select the right CPU, FPU, chipset, firmware, memory etc. to simulate one of those faster classic machines would be close enough to the real thing to not notice a difference.

Having a super fast Amiga in a way that the classic machines never were is not the first priority of MiSTer IMO, very accurate implementations of slower hardware comes first. This not quite there yet with the Amiga 500 on the MiniMig core (I submitted a ticket for graphics problems with Hybris months ago, it races the beam and doesn't work correctly on MiSTer).

If I'm not mistaking the 68020 in the MiniMig core is not cycle-exact and runs at a faster speed than one in an Amiga 1200.

[limi]

I agree that cycle-exact implementations would probably not be noticeable with these faster machines. IMO being able to select the right CPU, FPU, chipset, firmware, memory etc. to simulate one of those faster classic machines would be close enough to the real thing to not notice a difference.

Exactly.

Having a super fast Amiga in a way that the classic machines never were is not the first priority of MiSTer IMO, very accurate implementations of slower hardware comes first. This not quite there yet with the Amiga 500 on the MiniMig core (I submitted a ticket for graphics problems with Hybris months ago, it races the beam and doesn't work correctly on MiSTer).

Yeah, the Minimig core still has some issues, although a large class of them got eliminated recently when all the Graftgold games were fixed. Now it’s mostly smaller graphics corruption issues with certain games — they all run now, at least.

If I'm not mistaking the 68020 in the MiniMig core is not cycle-exact and runs at a faster speed than one in an Amiga 1200.

Correct.

[LamerDeluxe]

I didn't know the Craftgold games had problems before. I recently played Fire&Ice again and love Paradroid '90.

[kolla]

That's quite a bold statement: the only useful hardware config.

The only useful hardware config *missing*! Just about all configs there are working and useful solutions for, but fast CPU with fast, latency free, CPU independent native chipset... that is missing.

[robinsonb5]

I didn't know the Craftgold games had problems before. I recently played Fire&Ice again and love Paradroid '90.

It was a TG68 bug (stack frame related, I believe) which only happened with the TG68 in '020 mode. Slingshot fixed it on MiST some months back and it filtered through to MiSTer shortly afterwards.

[Bas]

So nobody here misses a CPU with MMU capabilities? That's at least my main takeaway from this thread so far.

[ByteMavericks]

The only thing I really used the MMU for was for Enforcer.

A faster CPU in the core would be awesome, however I'm not aware of other 68k implementations in FPGA that could give a better result - apart from the vampire 68080, and that's not likely to be coming any time soon I guess!

Of course, I could just be parading my ignorance around implementations there

[kolla]

So nobody here misses a CPU with MMU capabilities? That's at least my main takeaway from this thread so far.

I don’t miss it because I have that with software CPU emulation, I don’t see much point in implementing it on FPGA. The memory of the MiSTer is already quite different than old “classic” fast page RAM, so it is doubtful you can even do in a backward compatible way. And in any case, the simplified 040 MMU would be preferred over the apparently overcomplicated 68851/68030 MMU, so no AMIX...

There was long discussions on this topic on Apollo Core many years ago, nothing has changed since and MiSTer isn’t much different in this regard. If anyone think they know better and wish to prove naysayers wrong, then please do