Could such a thing exist? Amiga accelerator, FPGA.

HonestFlames

New member
Joined
Aug 18, 2010
Posts
346
Country
England
Region
Shropshire
With the Natami project claiming a 68k core implemented within their FPGA, what (apart from economic sense) would stop someone releasing a new accelerator board for A1200/A4000/A3000 which pretends to be e.g. a 68030, but is actually an FPGA module?

I mean, such a thing should run cool and possibly scale fairly well (200MHz equivalent 68030?).

If I were a hardware hacker, I'd certainly be looking into the possibility.

Just me, going barmy?
 

AndyLandy

Acceleration!
Treasurer
Joined
Aug 29, 2009
Posts
9,898
Country
UK
Region
Southampton
Been wondering that myself recently.

I believe there's a softcore 68000 implementation, but don't think there's anything higher than that.

A softcore 68020 would be fantastic if it could be done, particulary if you could start clocking it in the hundreds of MHz. If you could provide suitable glue-logic on a card, I don't think there's any technical reason it couldn't work.
 

HonestFlames

New member
Joined
Aug 18, 2010
Posts
346
Country
England
Region
Shropshire
I think it's just a problem of demand. There are so few of us who actually care enough to want such a device that the only way one will ever be developed is as a one-off by some electronics engineering student as part of their final project!

Of course, the argument is that we should all move on... but that kind of spoils the whole point :(
 

desiv

work in progress...
Joined
Oct 7, 2009
Posts
902
Country
USA
Region
Salem Oregon
There is a discussion about amikit looking for a project or the 1200 next on EAB.

The FPGA accelerator has come up..

It seems that people are saying that the fastest FPGAs that are cost effective are only fast enough to compete with 68020s.
The Natami people apparently have a softcore 68020.
The FPGA Arcade/minimig people added AGA to a 68000 softcore at 7Mhz and a lot of AGA programs ran (didn't use 68020 features/commands) and it ran faster than a 1200 for many things, although a lot of that might be the whole design of the system????

However, FPGAs will only get faster and cost less, so I would think if you design it now and you can get to (maybe???) a 20Mhz 68020 with FAST FAST memory, that would be very appealing and might compare favorably to faster real CPUs.

Then, a few years after that, as prices have dropped, you might have a model 2 (probably not upgradeable) that can do that (or 68030 or higher) at 100+ Mhz speeds.

Would be interesting...

I would think, if you could get it out at a reasonable price (?????) even a 68020 clocked higher than 14Mhz, but with 128M RAM or more would sell..

But who knows..

desiv
 

Cloudane

New member
AmiBayer
Joined
Aug 18, 2010
Posts
77
Country
UK
Region
Cumbria
I think it's just a problem of demand. There are so few of us who actually care enough to want such a device

That's the story of Amiga's life from about 1996 onwards ;)

It awes me (in a good way) that things like the sam440 and AmigaOS 4.1 are still going. Few they may be, but never underestimate the passion of Amiga enthusiasts!
 

Zetr0

Ya' Like it Retr0?
Joined
Nov 22, 2007
Posts
9,900
Country
UK
Region
Norfolk
I would love (and indeed dream) of a SoftCore CPU upgrade for my Amiga.


Technically:

You would need a CPU to LOAD the core to the FPGA on COLD boot-up.

This could be done with either clever bus arbitration and using the onboard 68k/020/030 CPU (pending machine and model) or with an Arm or other MCU controller.

This would cause a subtle initiall boot delay but on a reset this wouldn't be a problem.

There are many advantages to FPGA based 68k clone, mainly the size of the cache' (upto 1MB or 2MB)! as well as the optional extras for Network/Sound/DDR Memory Controller/LCD Driver etc.

But for me would be the fun in playing with the core of the CPU, refining it, adding extra ALU's to the design and then testing it within a live environment!!!


Unfortunately FPGA's that can contain an Moto-020 core are expensive and there are still legalities of copyright in regards to freescale that own the CPU circuits of the motorola, so as a shame it will be - its a year or two from here before its financially viable.

Although I would love to have fun with the 68000 core on an A500/+ or A2000 that would be a lot of fun indeed.


I think for todays use an FPGA could be used as a companion CPU for somthing thats (almost) a motorol - a Cold Fire v4 chip.

the Coldfire V4 has a lot of amazing features embeded in the CPU like Memory, UART, USB, LCD Driver, lots and lots. its also a refined 68000 core at anything from 30- 100MHz (and I have read faster) -

Alas to get it working with all the stuff in, stuff had to go, there is only so much realestate on the silicon... Unfortunately a few instructions are missing from the chip and have to be - reinturpreted, as such it breaks the compatability with the Amiga, infact quite a bit.

there was talk of opticode (cpu instructions) error traps where by these opticodes would be decoded, and then recoded as instructions that the cold-fire could understand... alas as the cold fire chip would be running the show, having to decode the opticode and then encode it again for it to process it... well lets say-- crippled the CPU...

Now that FPGA's are relatively cheap for small projects, it might be possible to consider a $30 component to infact do all this opticode repackaging for the cold fire, so as far as the cold fire goes its transparent - as it only gets the opticodes it can deal with.

Although this really does come down to CPU STACK and FRAME sizes, but I can say it would be an awefull lot of fun to find out if its possible.
 

Hell_Labs

New member
Banned
Joined
Mar 27, 2010
Posts
674
Country
England
Region
St.Helens, Lancashire
The Natami people apparently have a softcore 68020.

Ahahahahahaha. No. Don't ya think using a 060 on the dev boards and moving down to 020s would be a bit weird? Right now the bits of N68050 that actually work fookin crush the 060. Do people not realize that fpgas are better than they were 10 years ago or what...:blink: 200,000 LEs can fit a fast 68k with ease.
 

desiv

work in progress...
Joined
Oct 7, 2009
Posts
902
Country
USA
Region
Salem Oregon
The Natami people apparently have a softcore 68020.

Ahahahahahaha. No. Don't ya think using a 060 on the dev boards and moving down to 020s would be a bit weird? Right now the bits of N68050 that actually work fookin crush the 060. Do people not realize that fpgas are better than they were 10 years ago or what...:blink: 200,000 LEs can fit a fast 68k with ease.
OK, so you've laughed at my lack of technical knowledge on the subject. Fair enough.. :blink::roll::LOL::LOL:

And you've hinted that it might be possible...

But you haven't really answered the question posed..:Doh:

Since you know.. :bowdown:

What would be feasible in an FPGA as an accelerator for an A1200 today?

Would it be cost effective to produce an FPGA accelerator as fast as a 68040 or better?

You mention N6050.. Is that a softcore 68050 in the works? What speeds are we talking (SWAGs are fine)? How close is it to something that could be used?

Thanx,

desiv
 

Hell_Labs

New member
Banned
Joined
Mar 27, 2010
Posts
674
Country
England
Region
St.Helens, Lancashire
Okay so:

fpga a1200 accelerator pros:

really fast. REALLY fast.

easy to bugfix, just reflash.

made of all new parts.

cons:

really, really expensive.


As regarding the 050, it's actually superior to the 060 but with a few bits missing, so they gave it a lower name. Doesn't really fit in the naming scheme though. And I'm not allowed to say how fast it is beyond "faster".
 

AndyLandy

Acceleration!
Treasurer
Joined
Aug 29, 2009
Posts
9,898
Country
UK
Region
Southampton
Ahahahahahaha. No. Don't ya think using a 060 on the dev boards and moving down to 020s would be a bit weird?

Not in the slightest. The '040 and '060 CPUs are known for not being as compatible as the earlier CPUs. The reason is because the architecture was changed in the interests of cranking the clock speed higher. A softcore CPU takes that problem away. If you could clock an FPGA to 500MHz, you might as well use the most-compatible CPU design, which is the 68020. (Ignoring the 68000 and 68010, since they don't have 32-bit addressing). A 500MHz 68020 would utterly wipe the floor with a standard '060.

This was the same mentality that Apple used when they migrated to PowerPC and needed 68k emulation. In the first PPC-compatible OSes, there's an Easter Egg tribute to Gary Davidian, the guy who wrote the 68k emulation. It quotes "Make it look like an '020, make it run like a Cray"
 

Hell_Labs

New member
Banned
Joined
Mar 27, 2010
Posts
674
Country
England
Region
St.Helens, Lancashire
But apple's 68k emulation sucked. People even made replacements for it. And 500mhz is on the high side for a hardcopy asic, so lol.
 

HonestFlames

New member
Joined
Aug 18, 2010
Posts
346
Country
England
Region
Shropshire
Ahahahahahaha. No. Don't ya think using a 060 on the dev boards and moving down to 020s would be a bit weird?

Not in the slightest. The '040 and '060 CPUs are known for not being as compatible as the earlier CPUs. The reason is because the architecture was changed in the interests of cranking the clock speed higher. A softcore CPU takes that problem away. If you could clock an FPGA to 500MHz, you might as well use the most-compatible CPU design, which is the 68020. (Ignoring the 68000 and 68010, since they don't have 32-bit addressing). A 500MHz 68020 would utterly wipe the floor with a standard '060.

This was the same mentality that Apple used when they migrated to PowerPC and needed 68k emulation. In the first PPC-compatible OSes, there's an Easter Egg tribute to Gary Davidian, the guy who wrote the 68k emulation. It quotes "Make it look like an '020, make it run like a Cray"

It doesn't all come down to straight MHz. FPGA implementations of e.g. an '020 might be able to get rid of a whole bunch of wasted cycles where memory accesses are involved. Not to mention making each instruction the equivalent of one clock cycle, rather than the multi-varied timings a real '020 has. There are bound to be many other efficiencies too.

Perhaps powerful FPGA's are going to be expensive. As expensive as current second-hand Blizzard 060 boards though? Maybe. More expensive? Possibly. If the thing was blazing fast but with perfect compatibility, I'd save my pennies up to buy one.
 

Merlin

Ministry of Retr0bright and Street Judge
VIP
Joined
Nov 24, 2007
Posts
15,597
Country
UK
Region
Manchester
The big problem is going to be reverse-engineering all of the glue logic and interfacing code that was programmed into the GAL chips on the Blizzards, Typhoons, Cyberstorms etc. It was/is all proprietary and if I recall correctly, Jens Schoenfeld holds a lot of the information and data files and it isn't likely to get released.

It needs a reverse-engineering posse, with maybe a bounty being put up by those who really want a new accelerator produced; this is the most probable way that you would attract the talent required for the reverse engineering job in order to re-engineer the code content of the GAL chips for the new accelerator.

That's my two pence on the matter anyway.
 

Jumping Anaconda

Member
AmiBayer
Joined
Dec 31, 2009
Posts
683
Country
England
Region
Hertfordshire
If the likelihood is that faster FPGA's will arrive in the coming years, and that it would take a year or two to get the implementation working, would it not make sense to get it written now in time for the new hardware?

If there was the possibility of upgrading my Amiga that gave it the speed to browse the internet, and with a board that could provide network connections, I would definitely be interested in using my Amiga as a practical machine again. I think that might be the one genuine route for continuing the platform. I have always thought the PowerPC route was a dead end. If you had 68020 compatibility at very fast speeds, I could see people coming back to write for the machine, and AmigaOS 4.1 being dropped or ported back to 020. Either that or us all just settling on 3.9.

---------- Post added at 13:20 ---------- Previous post was at 13:10 ----------

BTW, what are the current costs for the kind of FPGAs that would be needed for this kind of project?
 

HonestFlames

New member
Joined
Aug 18, 2010
Posts
346
Country
England
Region
Shropshire
I do think there's going to be bigger demand for a new accelerator. Nobody has new 060 boards. I don't know if supplies of the CPU are drying up, or it's just no longer commercially viable to manufacture such niche Amiga hardware.

If the Natami project has developed a super-fast 68k core on a higher-end FPGA, there is perhaps a chance of a modestly-fast 68k core on a more affordable FPGA.

Maybe :tumble
 

Merlin

Ministry of Retr0bright and Street Judge
VIP
Joined
Nov 24, 2007
Posts
15,597
Country
UK
Region
Manchester
Commercially viable? Unlikely, since the production runs would be so relatively small that they would never recoup the setup costs.

Viable for the community to do this? Definitely possible. There are companies out in Asia that are prepared to make small batch run prototype boards at a reasonable cost. 68xxx chips are still available in some quantities although they are getting more scarce all of the time. FPGAs can be had at reasonable costs as long as you aren't trying to do too much with them, as the number of gates available on an FPGA is proportional to the cost.

There are a few discussions along these lines over at EAB although they tend to be negative and dismissive about such a project becoming a reality.

That's also what they said about Minimig though, so they could be wrong....
 
Last edited:

Hell_Labs

New member
Banned
Joined
Mar 27, 2010
Posts
674
Country
England
Region
St.Helens, Lancashire
Just go the whole bloody way and support the natami then, thomas is working on the next revision right now.
 

HonestFlames

New member
Joined
Aug 18, 2010
Posts
346
Country
England
Region
Shropshire
I plan on supporting Natami, though not in any other capacity than as paying customer and advocate (once I own one and can attest to its superbness).
 

Jumping Anaconda

Member
AmiBayer
Joined
Dec 31, 2009
Posts
683
Country
England
Region
Hertfordshire
I would have thought NatAmi was a much more adventurous project than a FPGA based accelerator/expansion board. Would I be misguided in that respect?
 

Buzzfuzz

Amiga tweaker
AmiBayer
Joined
Apr 16, 2010
Posts
1,708
Country
Netherlands
Region
Zuid Holland
That's also what they said about Minimig though, so they could be wrong....

Yeah Dennis pressed on and got it working and now units are sold, not much, but it is getting more attention.

There is high demand for highend turbocards and scandoublers, so why not combine it all on one card ?
If you indeed add modern sata and network to it, everyone would want one, whether is for an A1200, A4000 or an A500, A2000/3000.
 
Top Bottom