• Updated 2023-07-12: Hello, Guest! Welcome back, and be sure to check out this follow-up post about our outage a week or so ago.

68060 accelerator cards for Mac: Would you be willing?

Gorgonops

Moderator
Staff member
Wow, this is a fun zombie. You've got to love the "tech threads" that have such a high speculation-to-action ratio... ;^)

It's all been hashed to death, of course, but here's a really ignorant take on the "could we run MacOS on Coldfire?" question... why not just do it, for crying out loud? It seems that this thread is just hopelessly bogged down with minutia such as how one could somehow adapt the Coldfire CPU so it could physically be used as an upgrade CPU for an existing Mac, which is putting the cart *way ahead* of the horse. Why not prove out the concept on existing hardware (like a PCI slot-equipped Coldfire demo board) first? That's what the Atari Coldfire people did, and got a version of the Atari TOS patched and booting on a demo board well before any of their custom hardware was available.

It's been mentioned in one of these threads before, but one of the beautiful things about the classic MacOS when thinking about a project like this is its level of device independence. Emulators such as BasiliskII run MacOS in a virtual environment which has very little in common with a real 68k Macintosh simply by patching the ROM image appropriately. It seems to me that the simplest way to do Mac-On-Coldfire would be to incorporate the "direct-execution" core of BasiliskII (as utilized on Amiga and other 68k platforms) into a "New World 68k" bootloader/kernel, incorporating the CK68Klib binary compatibility layer and with low-level drivers for disk and I/O devices borrowed from Linux/uClinux sources. All the software tinkertoys are there, and I imagine a competent and very knowledgeable hacker (IE, one of the maintainers of BasiliskII or Sheepshaver) could put them together in a few man-months of work.

The real question is whether the end result would really be workable. There are certain edge-case conditions of existing 680x0 code that CK68Klib doesn't handle, and whether MacOS uses them and if so whether it's practical to patch around them are all open questions. But clearly it should be possible to deduce the answers before running off deep into the fantasy land of constructing a 680x0 socket-compatible accelerator for existing Macs.

A suitable Coldfire demo board isn't cheap, mind you. You're probably looking at a couple thousand bucks. Ironically a better approach might be to use an Atari Coldfire board. It's not cheap either, at 599 Euros, and availability is a problem, but it might be the best game in town. Of course, it's probably worth pondering that fact for a while. Any product based on this "research", whether it be an accelerator board or a complete new PCI-equipped Coldfire psuedo-Mac, is unlikely to end up much cheaper than that.

$600-ish dollars is a lot of money for a 300Mhz computer. Such a thing might well run 68K Mac programs at speeds comparable to a 1Ghz+ G4 or 2Ghz Pentium running an emulator, but... either of those options costs a heck of a lot less. (I can't find any benchmarks, but I'd be *seriously* surprised if the Atari Coldfire boards run EmuTOS faster than even a semi-modern PC running ARAnyM.) So clearly, the *only* reason to do this would be for love. Are there really enough people out there with undying love and nostalgia for *specifically* 68K Macs that isn't satisfied by emulators to support such a project? I have no idea. The Amiga and Atari people were faced with the complete extinction of their platforms so sketchy and uneconomical low-volume-production hardware wizardry was the only option the hardcore fans had to keep their love alive. Macs didn't die, they evolved, so... is there a point? Eh, who knows. All I know is there's an awful lot of talking and very little action. ;^b

 

johnklos

Well-known member
$600-ish dollars is a lot of money for a 300Mhz computer. Such a thing might well run 68K Mac programs at speeds comparable to a 1Ghz+ G4 or 2Ghz Pentium running an emulator, but... either of those options costs a heck of a lot less. (I can't find any benchmarks, but I'd be *seriously* surprised if the Atari Coldfire boards run EmuTOS faster than even a semi-modern PC running ARAnyM.) So clearly, the *only* reason to do this would be for love. Are there really enough people out there with undying love and nostalgia for *specifically* 68K Macs that isn't satisfied by emulators to support such a project? I have no idea. The Amiga and Atari people were faced with the complete extinction of their platforms so sketchy and uneconomical low-volume-production hardware wizardry was the only option the hardcore fans had to keep their love alive. Macs didn't die, they evolved, so... is there a point? Eh, who knows. All I know is there's an awful lot of talking and very little action. ;^b
If Ataris had meaningful preemptive multitasking, then a ShapeShifter / Basilisk type emulator / virtualizer would have already been made. There were ways to run MacOS on Ataris, but as far as I know they all required MacOS to entirely take over the whole OS, so they wouldn't work on the Coldfire board without lots of modification. But the Atari Coldfire-on-a-PCI board is an open project, so all documentation is available, so it's possible for someone with money and time.

Regarding benchmarks, emulators with JIT caching can run noticeably faster than a real m68060, but if you want or need to emulate an MMU, you can't use JIT, so the speed of a very fast x86 is not necessarily faster than a real m68060. I've heard from NetBSD Amiga developers with access to m68060 systems that some things are faster and some are slower - UAE isn't faster overall, but of course a 2+ GHz AMD or Intel system is much easier to find or buy than an Amiga with an m68060.

 

Gorgonops

Moderator
Staff member
Regarding benchmarks, emulators with JIT caching can run noticeably faster than a real m68060, but if you want or need to emulate an MMU, you can't use JIT, so the speed of a very fast x86 is not necessarily faster than a real m68060...
Emulating an MMU is a non-issue for MacOS so it's perfectly fair to note that even a fairly wimpy PC will run it far, far faster (on the order of hundreds of time faster, for certain tasks) than any Coldfire or 68060 box will be able to. If an MMU were actually vital for running Atari, Amiga, *or* MacOS software then I'm sure someone would be able to get JIT+MMU working together. It's just not a burning issue for anyone not running Linux or BSD, which is the vast majority of the users of these emulators.

(After all, if you want to run Linux or BSD why not just run it on the host? There's nothing particularly special about NetBSD on a Mac or Amiga other than the fact it runs on a Mac or Amiga.)

As an aside, I actually did stumble across a foreign site which had a collection of benchmarks comparing various "real" Atari computers to ARAnyM, and one of the things listed was a Coldfire demo board. It's interesting to note that the 200Mhz Coldfire barely beat the 100Mhz 68060. So... more reason to regard the whole enterprise as pointless?

 

John Hokanson Jr.

Well-known member
THE THREAD LIVES!

Just out of curiosity, as a means of settling whether this is even a worthy idea... what was the point in which 68k emulation on PPC overtook 68k code running natively on a 68040?

Ignore for a moment the existance of FAT binaries or recompiling for PPC.

I know the 601s were pretty crap at this (it *is* emulated), but were the 603s or 604s able to match the speed of a real 68040?

Given that Apple wrote a pretty decent emulator, I suspect a 68060, or even Coldfire, would fill only a VERY small need. Thus, the first person that responded to this thread (aeons ago) probably had the most beautifully simplistic answer. Which is to get a cheap G3 or possibly even a lesser PPC.

 

Bunsen

Admin-Witchfinder-General
Considering you can pick up a G3 iMac or a Beige for $20-$50, I'd say that bird done flown.

 

John Hokanson Jr.

Well-known member
Right, but you can bet that bird will try to rise from the dead at some point. There have already been several 68060 topics on 68kmla. It would be nice to have a full accounting of why and WHEN it doesn't make much sense.

BTW, with a certain amount of amusement, I note that the later Amiga accelerators used PPC 601 and 603 chips. Kinda makes one wonder.

 

Bunsen

Admin-Witchfinder-General
I don't think the 601s were even that "crap" at it, certainly not the faster ones in the 7100/8100

 

Unknown_K

Well-known member
Those Amiga 601/603 accelerators had a 68K stuck on them for 68K (68060 mostly) code and Amiga OS was modified to support PPC and the apps had to be PPC (very few).

 

trag

Well-known member
I don't think the 601s were even that "crap" at it, certainly not the faster ones in the 7100/8100
The original 68K emulator in PPC Mac ROMs was an interpreting emulator and even in a loop, it would reinterpret the instructions over and over. This cost the emulator a lot of performance.

Connectix wrote a caching interpreter which would keep chunks of the interpreted code and reuse them as needed, thus vastly speeding up loop structures and other repetitive chunks of code.

I think that with Connectix's Speed Doubler installed, an 80MHz PPC601 was pretty comparable to the fastest 68040.

In the PCI PowerMacs, Apple copied (or licensed?) Connectix's code and used that as the built-in 68K interpreter so a huge performance boost was seen beyond the improvements brought with faster CPUs, and more efficient memory subsystems.

If you're running a NuBus PPC or a PPC upgrade in a 68K machine, get Speed Doubler and install it. If you're using a PCI PowerMac there's little need for Speed Doubler (though some of the disk management/copy enhancements are nice).

 

johnklos

Well-known member
I don't think the 601s were even that "crap" at it, certainly not the faster ones in the 7100/8100
What trag said, plus the more significant speed differences were attributable to the larger caches in the newer PowerPC chips. One interesting tidbit is that there were some low end machines with the 603 (without the e at the end) which were complete dogs running m68k code because the m68k emulator code couldn't all fit in the 603's caches. The 603e was quickly released to address that issue.

 

John Hokanson Jr.

Well-known member
So, in a nutshell, there is really no practical reason for such an "upgrade." SpeedDoubler sufficiently boosts 68k emulation on 601s and 603s, and the 603e acheives parity with no additional software.

Thus, a 68060 or Coldfire Mac would only be for bragging rights and the lulz.

About right?

 

Concorde1993

Well-known member
Thus, a 68060 or Coldfire Mac would only be for bragging rights and the lulz.

About right?
From what I've read on this thread, John, I believe that's the message the majority of the members are trying to send out. If the cost of the upgrade were not as expensive as, say, picking up a couple of G3 units, then perhaps there would be a greater demand for this accelerator card. One also has to look at the availability of the card as well.

Not all mods are meant to be.

 

trag

Well-known member
So, in a nutshell, there is really no practical reason for such an "upgrade." SpeedDoubler sufficiently boosts 68k emulation on 601s and 603s, and the 603e acheives parity with no additional software.
Thus, a 68060 or Coldfire Mac would only be for bragging rights and the lulz.

About right?
Nope. You're kind of missing the point. But that's okay. It's well buried and partly composted.

Sure, you can get a newer faster PPC based machine. But it won't run the earlier OS and software.

The point behind speculating about a 68K upgrade is to take older machines, and the OS and software support those older machines provide, and make it much much faster. Preferably in the old form factor as well.

Is there much practical need for that? Probably not. But some folks are attached to the looks of older machines and some folks prefer the older software or just like to run it from time to time, and in many cases it would be nice if that older software on those old adorable machines would run much much faster.

 

John Hokanson Jr.

Well-known member
You are presumably referring to having a ridiculously fast System 6 machine. Because I was under the impression most PPCs before the G3 will run System 7 just fine (and do so faster).

I suppose it would be neat, but I just don't see anyone investing the time or money to actually do it.

 
Top