FPGA based A500 accelerator

[kipper2k]

Originally Posted by thgill

Wouldn’t mind seeing an A1200 version of this card.

It is on the list, not too long a list as the design is very portable

[thgill]

Fantastic!

[khph_re]

Assuming an A1200 version turns up at some point, how does 'double the speed of a 40 MHz A4000.' compare to a base 060? And is there any special 060 code that would stop demo scene stuff running on 020 instructions?

[Mathesar]

Originally Posted by kipper2k

We have quite a few more features to go yet... I think i know why the bootable SD card has not been done before, there was a lot of reading and some misinformation too.

A minimig can boot from SD. It also implements a Gayle style ide controller and has some software running on a microcontroller to emulate an ide drive. I guess this is something similar? Still very cool though

[Mathesar]

Originally Posted by khph_re

Assuming an A1200 version turns up at some point, how does 'double the speed of a 40 MHz A4000.' compare to a base 060? And is there any special 060 code that would stop demo scene stuff running on 020 instructions?

I think A4000/40 means a 25MHz 040 in AIBB, not a 40MHz 040.
That would mean this accelerator is about on par with a 50MHz 060 in this test?
(Pretty good!)

[kipper2k]

@mathesar The reference to the SD booting was in reference to the Amiga itself not requiring a CF card/HDD to boot the Amiga from. The minimig is a standalone device and a different species so to speak.

@khph_re There are 060 instructions not included in the 020 so some programs cannot run. Mike was referring to the imath score in comparison to the 40MHz A4000 on his blog . I dont think he was making a general A4000 performance evaluation

[nikosidis]

Before when I used Amiga Emulators I mostly ran 020 as fast as possible. That did make most 060 demos run nicely.

Now I only use real hardware. I will buy this fpga for my A1200 if not too expensive.

[kipper2k]

Originally Posted by nikosidis

Before when I used Amiga Emulators I mostly ran 020 as fast as possible. That did make most 060 demos run nicely.

Now I only use real hardware. I will buy this fpga for my A1200 if not too expensive.

Mike has posted on his initial blog that expected price will be ~200 Euro, there are more features in planning also

[nikosidis]

That seams reasonable 450 for Vampire A1200 and they have some real competition.

Personally I'm only interested in AGA and it should be possible to run quite a few 060 demos at full speed.

ScummVM AGA will also be nice with this.

[robinsonb5]

Originally Posted by kipper2k

@khph_re There are 060 instructions not included in the 020 so some programs cannot run. Mike was referring to the imath score in comparison to the 40MHz A4000 on his blog . I dont think he was making a general A4000 performance evaluation

The way AIBB presents its benchmark scores is a bit confusing: the 25 in "A3000-25" refers to the speed, 25MHz, but the the "-40" in "A4000-40" refers to the CPU '040, which is still 25Mhz! Also, the numbers shown beside each system show that system's speed compared with the currently selected reference system (A3000-25 by default) and don't show anything relating to the speed of the system under test - the only field that changes depending on the system being tested is the one marked "You".


The performance of this card is looking like it will be excellent, but expect it to feel more like "the fastest '030 ever" than anything in '060 territory.

[ex68k]

Originally Posted by robinsonb5

The performance of this card is looking like it will be excellent, but expect it to feel more like "the fastest '030 ever" than anything in '060 territory.

"The fastest '30 ever" is pretty good, definitely enough for a lot of things.
Than add FPU & MMU and there you go ;-)

[kipper2k]

As Mike has stated, the performance numbers are expected to increase once the final board is done with a faster FPGA and trace length etc are more uniform. I am very optimistic about the final product

[ex68k]

Originally Posted by kipper2k

As Mike has stated, the performance numbers are expected to increase once the final board is done with a faster FPGA and trace length etc are more uniform. I am very optimistic about the final product

Anything faster then the cyclone 5?

[NovaCoder]

Originally Posted by kipper2k

As Mike has stated, the performance numbers are expected to increase once the final board is done with a faster FPGA and trace length etc are more uniform. I am very optimistic about the final product

Have you guys considering adding 040 & 060 Opcode support one day?

I've no idea how feasible it would be but I believe this is what the Vampire crew did (take the old MiniMig core and update it)

The FPGA would not need to emulate the more sophisticated architecture of the 040/060, it would just need to run the software.

Apparently lots of 060 software contains unsupported 040 opCodes so a core that supports the full feature set of both the 68040 and 68060 would run everything faster

[Lemaru]

If using a Gayle IDE controller implementation, have you a workaround to stop it interfering with the PCMCIA port? Other accelerators that use that stop you from being able to use PCMCIA which isn't ideal

[mkstr]

Originally Posted by Lemaru

If using a Gayle IDE controller implementation, have you a workaround to stop it interfering with the PCMCIA port? Other accelerators that use that stop you from being able to use PCMCIA which isn't ideal

Yes, as it currently stands any motherboard Gayle would be masked entirely, but I don't think it would be a huge problem to merge the two. It would potentially hit performance a tiny bit as it would slow the internal Gayle down to the speed of the motherboard one, but the actual IDE registers would be unimpeded. I guess there could be a runtime switch for it.

[robinsonb5]

Originally Posted by mkstr

Yes, as it currently stands any motherboard Gayle would be masked entirely, but I don't think it would be a huge problem to merge the two. It would potentially hit performance a tiny bit as it would slow the internal Gayle down to the speed of the motherboard one, but the actual IDE registers would be unimpeded. I guess there could be a runtime switch for it.

Have you considered the possibility of an autoconfig ROM-based driver - like how SCSI cards used to do it back in the day? Then you could leave the onboard Gayle fully functional when it comes to bringing the device to A600 and A1200 users?

[mkstr]

Originally Posted by robinsonb5

Have you considered the possibility of an autoconfig ROM-based driver - like how SCSI cards used to do it back in the day? Then you could leave the onboard Gayle fully functional when it comes to bringing the device to A600 and A1200 users?

Yes, not least because that would allow for a far more efficient interface into the SD controller including DMA, but it's a job for later.

[kipper2k]

There is a lot of things going on around us all atm with the Coronavirus pandemic so things will drop a gear or two while priorities shift at home. once things start to normalize then it will pick up again.

Everyone stay safe and protect us old geezers

[kipper2k]

Time for a quick update, Mike has ordered a testing board to test out some new theories so progress hasn't completely stopped, in the meanwhile i am keeping myself amused in the lockdown by trying to make something much bigger than previous projects