What would you buy for classic Amiga?

[amigadave]

Out of curiosity, how would this work?  I know this is basically how the IndivisionECS works, by replacing Denise with an FPGA (and from what I understand, the ECS model is actually more advanced than the IndivisionAGA, since the ECS can operate without the Denise chip entirely)...  but since most AGA chips are SMD, are you proposing something like a "super IndivisionAGA", that would graft onto the top of the AGA chips, hijack their signals, and enhance them?  Or something entirely different, that might fit in the video slot of big box Amiga's?  Just trying to wrap my brain around how this might work.  

Yes, I think that Jens has learned enough about how to replace signals from the original Amiga Custom chipset and several programmers have toyed with improvements to the AGA video output (either in working demonstrations, or in theory), to make such a device possible for all Amiga models, as well as new stand alone FPGA systems.  Ideally, it would be developed for all models, with some being clip-on style, and others could be designed for the big box video slot.

[SamuraiCrow]

Thanks for clearing up my mistake.  I did not realize that the 400MHz result was not from the board(s) the team is working on producing as accelerators for many Amiga computer models.

This changes my perception of what can be accomplished with the yet to be released accelerator boards, from being a "Game Changing" amount of performance increase, to simply a incremental performance step increase, above what is already available using old 68020, or 68030 CPU's, but not the 40MHz to 50MHz 68040 & 68060 CPU's.

How does a FPGA soft-core running at an equivalent of a 68000 @ 120MHz compare to a real 68060 CPU running at 50MHz to 100MHz  (which seems to be the fastest I remember any of the real 68060 chips being over-clocked to)?  I would assume that the FPGA soft-core 680x0 @ 120MHz would provide less performance than a real 68060 @ 50MHz, let alone an over-clocked 060 @ 100MHz.

Combining the fast soft-core design on a new motherboard with faster RAM & a faster local bus to all of the other components, plus the addition of on board USB, Ethernet & a SATA controller, will further increase over-all performance, but I think I will go back to dreaming of my FPGA Arcade Replay with the 68060 daughter board, as my preferred 68k Amiga clone.

With my perception of what is currently being worked on corrected, I think that more useful progress for 68k users could be made by improvements to the AGA chipset, instead of putting so much work into 680x0 soft-core designs.  At least until the performance and price of FPGA chips improves significantly, to allow the soft-core designs to exceed 68060 performance at a cost of only a couple hundred dollars.  Until that happens, I think that improvements to the AGA chipset performance and features (including higher resolutions and color depth choices, as well as the ability to access greater than 2mb of Chip RAM), by using FPGA chips, would be a better use of the talented FPGA programmers in our community.  Just my personal opinion, as the FPGA programmers will continue to work on what ever THEY think is most important.

FPGA accelerators will be nice, specially if they also include features like USB and/or Ethernet and additional RAM with faster bus speeds, but if they provide less performance than existing 68040 & 68060 accelerators, the new accelerators will be less impressive than I had first hoped.  Users who already have an 060 accelerator for their Commodore Amiga computers, probably also have a Deneb for USB and an Ethernet Zorro card, plus additional Fast RAM, so they will not replace a faster accelerator for a slower one.  The people who will want an Apollo FPGA accelerator are A500, CDTV, A600, (A1000 if it will fit) and some A2000 owners.  CD32 owners are desperate for a new accelerator design, but I don't know if the Apollo team will be able to shrink their design down enough to fit inside of the CD32, or if they can find the right connectors to make it work in a CD32.

Shame on you Gunnar for fooling me with that 400MHz test result and getting me all hyped up for something we won't see available for sale to most Amiga users for another 2 to 5 years (specifically an FPGA accelerator that can provide 400MHz performance).

The tests on the Apollo core website are all done on the same Arria board.  The efficiency of the Apollo core is very good.

A 120 MHz board will be faster than an overclock on a 68060 board by more than 20% and cost less as well.

::edit::
Also, a 400 MHz Raspberry Pi is not likely to be faster than this board except for OpenGL ES based 3D acceleration and even then, softcore add-ons are expected to come out for a budget-minded price after the initial release.  The hi-def AGA++ core, for example will address all 128 Megs of board memory as Chip RAM.  Can you tell me why a mere AGA chipset clone would be better?

[alphadec]

would be nice if a-eon would get on board and try to push through a amiga fpga-clone so the amiga community could have also more choices and also we dont forget the classic amiga.

[QuikSanz]

would be nice if a-eon would get on board and try to push through a amiga fpga-clone so the amiga community could have also more choices and also we dont forget the classic amiga.

Now that brings to mind a new model specific replacement Motherboard with modern features added.
Now that would be too cool!

Chris

[matthey]

The tests on the Apollo core website are all done on the same Arria board.  The efficiency of the Apollo core is very good.

If you are talking about the performance benchmarks here:

http://www.apollo-core.com/index.htm?page=performance

Then the fpga is a Stratix which is significantly faster and more expensive than the Arria. Some of the Arria fpgas are borderline affordable but generally the Cyclone series gives better bang for the buck and there are sizes that are large enough for everything an Amiga would need.

All the benchmarks here:

http://www.apollo-core.com/bringup/

show results for Majsta's accelerator with a tiny Cyclone II fpga. Some of the benchmark MHz numbers given are calculated by performance (some benchmark programs are not accurate).

A 120 MHz board will be faster than an overclock on a 68060 board by more than 20% and cost less as well.

You must be referring to the new sandwich accelerator with larger fpga. Yes, Phoenix in it should be better at integer performance than a 68060. Phoenix in Majsta's accelerator is probably more equivalent in performance to a 68040 at 120-160MHz. It's going to perform better with old 68000-68040 code than the 68060 which needs optimizations to fully take advantage of superscalar execution. Phoenix is handicapped severely by the small Cyclone II but it still approaches the integer performance of a 68060@50MHz and no doubt outperforms it in some areas.

would be nice if a-eon would get on board and try to push through a amiga fpga-clone so the amiga community could have also more choices and also we dont forget the classic amiga.

If A-Eon had invested $1 million in Natami and burning an ASIC based on the Phoenix fpga CPU core, the 68k wouldn't be the slow poke anymore. I bet an ASIC could outperform a SAM 440 while being compatible with most of the old 68k software. It's interesting that the problems with SMP (exec/ables.i macros) are more likely solvable in an fpga (or Phoenix+Amiga ASIC) than with a PPC based AmigaOS. It wouldn't be too difficult to add multi-threading and/or more cores (the hardware side but the software side could be tricky) with Phoenix. A lot of Amiga users like the 68k with compatibility and won't abandon it.

[amigadave]

The hi-def AGA++ core, for example will address all 128 Megs of board memory as Chip RAM.  Can you tell me why a mere AGA chipset clone would be better?

Who asked for "a mere AGA chipset clone"?  I brought up the topic of an improved AGA chipset clone, the so called SAGA, or Super AGA, chipset that some developers have worked on since the early days of the Natami project (perhaps before that, I don't know), because I thought that it would be more beneficial to have programmers work on completing and perfecting a new SAGA standard that would allow higher resolution and color depth software applications and games to be created for Amiga computers and clones equipped with a device which provides SAGA features.  My preference to have programmers work on SAGA instead of the soft-core 68k, was based only on my misunderstanding that a soft-core 68k CPU would only be faster than a real 68060 when loaded into an FPGA that costs thousands of dollars each.

I keep getting bounced back and forth by one side or the other, when one side says that the actual performance of the (relatively cheap) FPGA based accelerator boards will provide performance slower than a real 68060, while the other side comes back with statements that the bench mark test results are from the FPGA to be used in the accelerator, not some bigger, faster, FPGA that costs thousands of dollars to purchase.

At this point in time, I don't know what is real and what is fantasy, so I guess I will just wait until an actual product is available for purchase.  I had hoped that work on the design and firmware would be finished and production of the accelerator boards would have begun, so assembly and testing could begin, followed by sales to customers shortly afterwards.

[matthey]

My preference to have programmers work on SAGA instead of the soft-core 68k, was based only on my misunderstanding that a soft-core 68k CPU would only be faster than a real 68060 when loaded into an FPGA that costs thousands of dollars each.

I keep getting bounced back and forth by one side or the other, when one side says that the actual performance of the (relatively cheap) FPGA based accelerator boards will provide performance slower than a real 68060, while the other side comes back with statements that the bench mark test results are from the FPGA to be used in the accelerator, not some bigger, faster, FPGA that costs thousands of dollars to purchase.

A medium to large Cyclone V fpga that would be adequate for fpga 68k CPU+FPU+Amiga SAGA custom chips is <$50. The Phoenix sandwich accelerator mates with a standard Cyclone V fpga board that costs $50 (although the Cyclone V is small and doesn't have room for a full FPU). An Altera mid-range Arria fpga starts at about $50 and goes up in price from there. An Altera high-end Stratix fpga starts at several hundred dollars and goes up to thousands of dollars.

At this point in time, I don't know what is real and what is fantasy, so I guess I will just wait until an actual product is available for purchase.  I had hoped that work on the design and firmware would be finished and production of the accelerator boards would have begun, so assembly and testing could begin, followed by sales to customers shortly afterwards.

Majsta's Amiga 600 fpga based accelerator is for sale now. Most people are still using the TG68 in it but there are people testing the Phoenix fpga CPU. I believe Majsta is planning a little larger Cyclone III fpga based Amiga 500 accelerator and the Apollo Team is working on a more universal Amiga 500, 1000, 2000, CDTV based sandwich accelerator. I don't know which of the latter accelerators will be available first or when.

[HardStep]

Fpga accelerator for A1200, even if "only" matching performance of a 040/060, including faster than motherboard clock port(s). Sign me up for 2 of those:)

[utri007]

Graphics card for Zorro slot.

[matthey]

Graphics card for Zorro slot.

How about gfx card for CPU slot without the Zorro bottleneck?

[biggun]

The 400MHz figure was from a multi thousands of dollars Arria devboard.  The Cyclone 5 can hit 120 MHz or so.

SAM, you talk about two different thinks.

Dave was talking about a Sysinfo Screenshot which measured the Speed as the same as a 68000@350  MHz
And this result was actually done the $90 Vampire card.

Of course the Vampire is just an entry level card.

The Phoenix card we test right now - is about twice as strong.
E.g it would score like a 700 MHz 68000.

The high end card you refer to SAM would again result in twice the score.
So getting a performance equally to an 68000 over a giga herz is really possible.



Sam, what you mean were the benchmark results where our 68K did blow the PowerPC out of the water.
These scores were actually done on a high end FPGA card.
But these are not the scores Dave is talking about.

Dave I think you were talking about Sysinfo and AIBB numbers right?
These scores were done on the low priced Vampire.

[biggun]

I
If A-Eon had invested $1 million in Natami and burning an ASIC based on the Phoenix fpga CPU core, the 68k wouldn't be the slow poke anymore. I bet an ASIC could outperform a SAM 440 while being compatible with most of the old 68k software.

You do not need an ASIC to come in SAM performance.
Aven a quite small FPGA can outrun SAM in many benchmarks.


You also not even need a million for to build an ASIC.

[biggun]

Thanks for clearing up my mistake.  I did not realize that the 400MHz result was not from the board(s) the team is working on producing as accelerators for many Amiga computer models.

Guys please don't confuse Dave here.

The results you mean were from the low cost Vampire board $90.
The results form the also cheap Cyclone board are aboput twice as good.

This changes my perception of what can be accomplished with the yet to be released accelerator boards, from being a "Game Changing" amount of performance increase, to simply a incremental performance step increase, above what is already available using old 68020, or 68030 CPU's, but not the 40MHz to 50MHz 68040 & 68060 CPU's.

A $90 Vampire has about the speed of a 50-100 MHz 68060

A low priced Cyclone5 CPU upgrade we work can reach a performance
roughly compareable to a 200-300 MHz 68060 systems.
You can not easily say this in 1 number as every benchmakr is different.

How does a FPGA soft-core running at an equivalent of a 68000 @ 120MHz compare to a real 68060 CPU running at 50MHz to 100MHz  (which seems to be the fastest I remember any of the real 68060 chips being over-clocked to)?  I would assume that the FPGA soft-core 680x0 @ 120MHz would provide less performance than a real 68060 @ 50MHz, let alone an over-clocked 060 @ 100MHz.

The $90 Vampire outruns an 68060@50MHz in most benhmarks .
The $90 Vampire beats an 68060@100 MHz in some benchmarks.


The Cyclone 5 systems is about 2-3 times stronger.

A high priced FPGA can again be a couple times stronger.

[alphadec]

The Phoenix card we test right now - is about twice as strong.
E.g it would score like a 700 MHz 68000.
.

When will we see a stand-alone unit. ?

[biggun]

When will we see a stand-alone unit. ?

Our roadmap is:
*Vampire release (600/500)
*Phoenix CPU card release (500/1000/2000)
*Phoenix CPU card releases (1200/3000/4000)
* Standalone system

[alphadec]

Our roadmap is:
*Vampire release (600/500)
*Phoenix CPU card release (500/1000/2000)
*Phoenix CPU card releases (1200/3000/4000)
* Standalone system

So standalone system will be released somewhere around 2017.. ?