[Vampire] The AMIGA Future Is NOW! AROS!

[SamuraiCrow]

I just read the thread. We already have two hdl implementations of the AGA chipset, mine and the minimig derived one. We have extensions such as extra memory, RTG, AHI audio etc already.

I guess the game here is to add propriety extra features and build a new platform, but that potentially breaks back compatibility. For them it makes sense to open source that part of the design to encourage people to use those features. Personally I'm much more interested in a new, cycle accurate (and high performance compatible) open source CPU, which is what we are working towards.
Why not just compile AROS to cheap modern CPU platforms?
/Mike

Where do I start?

• AROS runs hosted on Linux on the RasPi already.  There's hardly any software that runs on ARM AROS.  Same goes for x86 and AMD64.
• Running UAE4ARM on a RasPi 3 is good enough to emulate an A1200 quite well already.
• The SAGA core implements the chunky and YUV modes as a third playfield by re-purposing a register in AGA that was intended for a greyscale monitor that was never produced.  It's not a separate RTG mode, it's all integrated.  THat's actually the main reason for AROS: new graphics drivers that work as one.
• Since the closed-source 68080 core performs as well per clock as a Core2 Solo, it'll keep up with the modern systems quite nicely!  Especially when compared to an ancient 68060 clocked at twice the clock speed!  Note:  The minimum clock speed of the '080 is 78 MHz already so I hope you can clock your 68060 board at a minimum of 144 MHz...  ;-)

[OlafS3]

I just read the thread. We already have two hdl implementations of the AGA chipset, mine and the minimig derived one. We have extensions such as extra memory, RTG, AHI audio etc already.

I guess the game here is to add propriety extra features and build a new platform, but that potentially breaks back compatibility. For them it makes sense to open source that part of the design to encourage people to use those features. Personally I'm much more interested in a new, cycle accurate (and high performance compatible) open source CPU, which is what we are working towards.
Why not just compile AROS to cheap modern CPU platforms?
/Mike

you can compile AROS almost for anything today including ARM, X86, ARM64 and even PPC. Thanks for hint...

[OlafS3]

I don't have any problem.
But I call the things by his name.
Amiga is hardware and his OS.
Others is "like".

in this case... why do you use a SAM?

[OlafS3]

it sounds like he dislikes competition...

[mikej]

Where do I start?[/LIST]

• AROS runs hosted on Linux on the RasPi already.  There's hardly any software that runs on ARM AROS.  Same goes for x86 and AMD64.

The CM3 module is emulating just the CPU - so it looks like a fast 68K

• Running UAE4ARM on a RasPi 3 is good enough to emulate an A1200 quite well already.

Running the chipset in the FPGA and the CPU in the ARM gives (in my view) the best of both worlds. And there will be a CM4 etc in future cheaply to speed up the CPU. It also has local memory and HDMI out obviously as well.

• The SAGA core implements the chunky and YUV modes as a third playfield by re-purposing a register in AGA that was intended for a greyscale monitor that was never produced.  It's not a separate RTG mode, it's all integrated.  THat's actually the main reason for AROS: new graphics drivers that work as one.

If this is documented then it would be trivial to add the additional modes to any of the current hardware FPGA platforms

• Since the closed-source 68080 core performs as well per clock as a Core2 Solo, it'll keep up with the modern systems quite nicely!  Especially when compared to an ancient 68060 clocked at twice the clock speed!  Note:  The minimum clock speed of the '080 is 78 MHz already so I hope you can clock your 68060 board at a minimum of 144 MHz...  ;-)

113MHz, but it has MMU and FPU.

[Iggy]

you might change your mind if this ever happens :

"But we actually want to reach more.
We know that we can put APOLLO in to an ASIC.
And we know that then we can compete and beat even G3/G4/.. latest PowerPC or ARM Asics in performance."

(citation : ) http://www.apollo-core.com/knowledge.php?b=1¬e=5768

Sounds remarkably clueless for someone that has produced a working core.
Current predictions of faster speeds in FPGA are extremely unlikely without adding cache to the processor.
Much higher speeds in an ASIC will require further re-engineering of the cpu.

Do you actually think you just turn up the clock speed of a cpu like a rheostat in an analog circuit?

A small team of developers is going to make a 68K derivative outperform 2GHz 64bit PPC and ARM cpus?  

"we know..."

REALLY?

When monkeys fly out of Gunnar's butt.

[wawrzon]

@mikej:

actually your plan to emulate solely the 68kcpu on another rchitecture and making it look to the system like a fast 68k is what i and few others proposed all along. the difference is, that i thought in this case of an amiga accelerator design.
your concept sounds very good nevertheless and imho is a good complementary alternative to apollo/vampire approach/proposal. i wish you a lot of luck with it and stand by watching. how it unfolds.

what concerns aros 68k its very easy tm make it boot on amiga hardware. so it shouldnt be a problem to boot it on your board. i think there are people doing this (kola?). aros is working with p96 drivers via wrapper. they need to be loaded via arosbootstrap.

[Iggy]

@ mikej

Sounds good Mike.
And the goals are realistic.
Hey, functional MMU and FPUs, what an idea!

Instead of focusing on features that aren't supported by existing software.

[OlafS3]

@Iggy

Gunnar never claimed to outperform 2 Ghz CPUs... he wants to beat them per Mhz. He is a CPU guy so it is his personal motivation obviously (next to other things)

Regarding CISC they say it would be possible to do one with 1.2 Ghz. If true I do not know

[SamuraiCrow]

Sounds remarkably clueless for someone that has produced a working core.
Current predictions of faster speeds in FPGA are extremely unlikely without adding cache to the processor.
Much higher speeds in an ASIC will require further re-engineering of the cpu.

Do you actually think you just turn up the clock speed of a cpu like a rheostat in an analog circuit?

A small team of developers is going to make a 68K derivative outperform 2GHz 64bit PPC and ARM cpus?  

"we know..."

REALLY?

When monkeys fly out of Gunnar's butt.

PPC and ARM are RISC machine punching bags.  Gunnar used to work on PowerPC cores at IBM so he should know.  It AMD64 that's the instruction set to beat.

As for caches in an FPGA, there are bigger and more expensive FPGA models than what we're using....

Looking forward to the butt monkeys in Germany.  

[kolla]

113MHz, but it has MMU and FPU.

Do you see any possibility to make an acc card for "real" amiga hardware using the RPi Compute module?

[kolla]

Gunnar used to work on PowerPC cores at IBM so he should know.

Yeah, he _used_ to. What happened?

[Iggy]

@Iggy

Gunnar never claimed to outperform 2 Ghz CPUs... he wants to beat them per Mhz. He is a CPU guy so it is his personal motivation obviously (next to other things)

Regarding CISC they say it would be possible to do one with 1.2 Ghz. If true I do not know

That is closer to reality. It will still require a lot of reworking.
The problem with "quotes" is that they are usually fragmentary and often taken out of context.

That "we know" comment being a good example.
Inferring "we know" an ASIC descendant of the 68080 will outperform high end PPC and ARM cpus (with no mention of a "per mhz" basis).

[wawrzon]

Hey, functional MMU and FPUs, what an idea!

as a comptent person you probably are aware that 68k mmu emulation wont work with jit. without jit emulated 68k cpu is slower than apollo core. that has been measured.

[mikej]

as a comptent person you probably are aware that 68k mmu emulation wont work with jit. without jit emulated 68k cpu is slower than apollo core. that has been measured.

I was talking about the real 68K here. What is possible on the quad core 1G ARM remains to be seen when it is just running CPU emulation.

I'm more interested in performance per $. A CM3 module is something like 30$ and has 1G RAM and HDMI out. It's tough to compete with that with the FPGA.

I started this looking at the zynq SOC chips which have a fast FPGA device and on-board ARM, but it's still more cost effective to strap the CM3 module onto a spartan7 FPGA or something similar.


http://www.fpgaarcade.com/punbb/viewtopic.php?id=1221

[mikej]

Do you see any possibility to make an acc card for "real" amiga hardware using the RPi Compute module?

This was discussed. I have no interest in doing it, the accelerator market is somewhat saturated. It's much easier to interface the compute module directly to the FPGA as you can control timing. To fit in a 68K socket you would need a bridge device so you are losing the cost benefit. Perhaps a zynq with embedded CPU core would be interesting, but they are quite pricey.

"We know that we can put APOLLO in to an ASIC."

I can put anything into an ASIC. My day job is working for a fables ASIC house that actually makes chips (specialist high performance 28nm CPUs amusingly). I keep meaning to run the TG68K CPU through the Cadence synthesiser and get a timing estimate - it would probably go at least 500Mhz, maybe more.

That doesn't mean it's going to happen though. Licence costs for the tools, cells and memory compilers are very steep, as are the production costs. We make high price devices so there is a business case. I don't see a fast 68K processor selling for more than a few$ - it's got to compete against Atmel etc who make really quite nice ARM based SOCs.

There is test, characterisation and a million other things to consider apart from having just the RTL.

NXP still make 68K devices. While making a few hobby FPGA designs is not going to interest them, they will aggressively protect their IP if you try to make a business out of it.

The vampire core is fun and seems to work well, but I don't believe it's going to take over the world (unless ARM was to vanish tomorrow).
/Mike