Apollo Maggie 3D Chip - Preview Demo

[Seiya]

now is mandatory the porting of Unigine Valley for Amiga

[jbenam]

Originally Posted by willemdrijver

That's a fair question and to be honest I have no idea at the moment.
The only thing I can offer right now is the 25 fps performance on the alpha v0.1 demo with 512x512 texture mapping on a 640x480 true-color display.
But there are daily optimisations now on Maggie, so I would expect the performance to get much better.
I will ask the 3D experts about some benchmarking figures.

Thank you for your answer and your work on the 3D core. It's not a simple matter to create a "GPU" in FPGA. Godspeed!

[Promilus]

@eXeler0
voodoo 2 was like 1,5 voodoo1 (iirc got 2 texture units instead of just one, but still one pixel unit) and voodoo3 was like 2x voodoo2. I doubt Maggie does anything beyond Voodoo1 but as it is tightly coupled with both softcore CPU, softcore chipset and memory controller it should work better despite lack of "processing power". That's also why Vampire does better than what you could've expect from just MIPS and FLOPS figures.

[eXeler0]

My point here is primarily that we are shown alpha version 0.1 of something running on an fairly low-mid range FPGA so i expect this to be the first baby steps and will be improved incrementally. Maybe give it some time to mature before asking for benchmarks relative to voodoo3 :-)
It took them a while to do the math co processor, didnt it ;-)

[Promilus]

we are shown alpha version 0.1 of something running on an fairly low-mid range FPGA

Which would be high end decade ago and holy grail 2 decades ago when Voodoo4 (!!) was rushed to the market just before 3dfx died. It's not like you can squeeze any decent 3d core and ogl compliant as well (on comparable level of late 90s tech) into 5CEA5 used by apollo team in their products.

It took them a while to do the math co processor, didnt it

Yes and it was truncated to 64bit just because full 80bit was hard to fit, and for V2 it was even dumbed down to 56bit.

Maybe give it some time to mature before asking for benchmarks relative to voodoo3

I'd be mighty impressed if they could reach that with next from the line (iirc 125k LE? or was it 110) but I'm pretty sure that's outside of the reach of most consumer-level FPGAs.

[Seiya]

seriously, there is no demand about 68k 3D acceleration games. Yes, there is something, but on 68k 3D acceleration is very useless because require a lot of power especially fpu. Watching latest Amiga productions, gpu acceleration is not more a priority.
For me it's a waste of time that could be used for developing quality games and applications.

[grelbfarlk]

Hey let's just walk it back a few steps and maybe like on some artificial benchmark it's almost as good as the Permedia 2 on the Cybervision PPC! Then they can claim it is the most bestestest 3d accelerator for the Amiga EVAR! The mammary speed is like better by 3X! So it's THREE TIMES FASTARLIER!

[NovaCoder]

Personally I don't really see the point....it's just a continuation of the AMMX mindset which has created a niche FPGA computer which happens to have backwards compatibility with old Amiga software.

Each to their own though, I'm sure some people will enjoy it.

[Promilus]

It's exactly that - expanding features for that niche of hobbyists owning apollo products. It won't change much on the market as 3D graphics was already there long time ago but it gives new tools for all those developers who weren't satisfied with just more sprites and colors with their SAGA. It is not a bad thing but there won't be much use for it. Most likely even less than AMMX.

[meynaf]

Is this a fully featured GPU able to do 3D but also shaders and video decode ? If not, I don't see the point.

[Promilus]

but also shaders and video decode

Not a chance.

I don't see the point.

I don't see the point of having programmable shader capability in conjunction with pentium-equivalent processor. Those titles which requires programmable shaders won't run on it anyway. Also - such GPU would eat all the LEs from FPGA leaving nothing for AC68080 and it would still not be enough.

[meynaf]

Originally Posted by Promilus

Not a chance.

Then i don't see the point.

Originally Posted by Promilus

I don't see the point of having programmable shader capability in conjunction with pentium-equivalent processor.

Actually, '080 is more powerful than pentium-equivalent processor. Not by very much, but still.

Originally Posted by Promilus

Those titles which requires programmable shaders won't run on it anyway.

That depends. Not only games can use them, other software can as well.
F.e. I think it would be fun to code 8-bit or even 16-bit machines emulators using GPU power for screen conversion.

Originally Posted by Promilus

Also - such GPU would eat all the LEs from FPGA leaving nothing for AC68080 and it would still not be enough.

In todays FPGAs, yes, probably. But in the future ?

[Promilus]

Actually, '080 is more powerful than pentium-equivalent processor.

Last pentium (p54c) was 233MHz which should give around 400MIPS when AC68080 has ~340. It is quite similar architecture, AC has great advantage in aspect of memory controller (shouldn't be that hard to understand DDR3 SDRAM is much faster than SDR SDRAM @ 66MHz, L2 cache on-board and memory controller on board (and not integrated). But FPGA lacks speed (core frequency).

That depends. Not only games can use them, other software can as well.

To do that you'd have to make yet another abstraction layer, driver etc. Create whole new subsystem and integrate with OS. Look how fast ARM on ZZ9000 progresses... yeah.. exactly.

But in the future

In no foreseeable future such FPGA would be easily and cheaply obtained and then there's that aspect of hundreds of programmable units (simple but still) which could give you a headache just trying to imagine how to implement on FPGA.

[eXeler0]

Lots of speculation going on here,
Apollo forum would be the best place to start for the basics

http://www.apollo-core.com/knowledge.php?b=8&note=38071

[meynaf]

Originally Posted by Promilus

Last pentium (p54c) was 233MHz which should give around 400MIPS when AC68080 has ~340. It is quite similar architecture, AC has great advantage in aspect of memory controller (shouldn't be that hard to understand DDR3 SDRAM is much faster than SDR SDRAM @ 66MHz, L2 cache on-board and memory controller on board (and not integrated). But FPGA lacks speed (core frequency).

233Mhz P5 won't give 400mips (i don't think it even reaches IPC of 1.5), and even if it did, "mips" does not mean anything as it has to execute more instructions than 68k to achieve same result.

Originally Posted by Promilus

To do that you'd have to make yet another abstraction layer, driver etc. Create whole new subsystem and integrate with OS. Look how fast ARM on ZZ9000 progresses... yeah.. exactly.

ARM isn't 68k.

Originally Posted by Promilus

In no foreseeable future such FPGA would be easily and cheaply obtained and then there's that aspect of hundreds of programmable units (simple but still) which could give you a headache just trying to imagine how to implement on FPGA.

Then it may happen in another way. As ARM cores get integrated into FPGAs, one day same could happen for GPUs (ARM SoCs already include them and more). Or some off-the-shelf standard chip could be put on next generation of boards so FPGA does not have to do more than the cpu.

[coder76]

Originally Posted by Daedalus

And that's my point. Requiring specially coded software immediately rules out any existing software that might run on it, and will mean new software developed for it won't work with existing Amiga 3D hardware.

Look at RTG for example. How much software uses RTG and can use the RTG modes of the Vampire cards? And how much uses the SAGA screenmodes directly?

Sometimes there's a good reason for these additional software layers.

But if you move to separate 3D hardware, you'll also lose the capabilities of the Amiga custom chipset, like copper, sprites, and dual playfield. With a powerful enough 3D unit, these are perhaps not needed, but it's still good to have them, for more simpler games. A mix of 2D and 3D capabilities of the display hardware is always better, especially with Amiga, since it's still very underpowered compared to modern hardware.



And as someone pointed out, it's also possible to abstract the Amiga custom chip hardware, like any other hardware. But with hardware interface also being clearly defined and not subject to continuous changes is also a clear advantage, as people can start using the hardware directly when it is available.

[Promilus]

@meynaf

ARM isn't 68k.

No it is not. Neither are ALU from GPU which you'd like use for program execution. But fact is - there already is SDK and there are ppl using ZZ9000 yet basically no such software exists using ARM. Why you assume there would be any need for software using programmable GPU units?

As ARM cores get integrated into FPGAs, one day same could happen for GPUs (ARM SoCs already include them and more)

That future is already here but that's not something Gunnar want to dig into. There already are FPGAs integrated with ARM and GPU and video codec (Zynq UltraScale+ EV for example) but it is not direction Apollo team want to lean into. And obviously FPGA itself which costs 300+ USD isn't something which would gather masses (also that's altera's competitor device and would require quite a lot of verilog code - hw specific - to rewrite).

233Mhz P5 won't give 400mips

Pentium 100MHz generates ~170DMIPS. 060@75MHz generates 110. Hypothetical 060@100 (OC) should give ~150. We know AC68080 is faster per clock (from both Pentium and 060) but it has lower clock than Pentium MMX. GPU with programmable shaders became only popular around ~1200MHz Pentium III and Athlon K7 so there is absolutely no chance to port such games to AC68080.

[Daedalus]

Originally Posted by coder76

But if you move to separate 3D hardware, you'll also lose the capabilities of the Amiga custom chipset, like copper, sprites, and dual playfield. With a powerful enough 3D unit, these are perhaps not needed, but it's still good to have them, for more simpler games.

Which is why 3D hardware always goes hand in hand with RTG hardware. RTG hardware will provide these 2D functions in a much more powerful and less limited way than using native chipset modes.

A mix of 2D and 3D capabilities of the display hardware is always better, especially with Amiga, since it's still very underpowered compared to modern hardware.

See above.

And as someone pointed out, it's also possible to abstract the Amiga custom chip hardware, like any other hardware.

Of course it can, and is pretty much essential for a feature like this to have any significant use. Like the RTG driver that's used by almost every Vampire user for their pretty Workbench screens, abstraction to an established standard allows existing software, existing development tools, existing development tutorials, existing porting guidelines to be used without modification.

But with hardware interface also being clearly defined and not subject to continuous changes is also a clear advantage, as people can start using the hardware directly when it is available.

That's not any different to any other 3D hardware. If you want to access the Permedia 2 chip on a BlizzardVision directly, you can do that. That hardware is much less likely to change than the Maggie chip, for example. But then your software would naturally only work on a Permedia 2-equipped machine.

[meynaf]

Originally Posted by Promilus

No it is not. Neither are ALU from GPU which you'd like use for program execution. But fact is - there already is SDK and there are ppl using ZZ9000 yet basically no such software exists using ARM. Why you assume there would be any need for software using programmable GPU units?

Good point. Indeed software for it would be problematic.
As a coder i would be more interested in developing for that than for ARM, though.

Originally Posted by Promilus

That future is already here but that's not something Gunnar want to dig into.

Too bad.

Originally Posted by Promilus

Pentium 100MHz generates ~170DMIPS. 060@75MHz generates 110. Hypothetical 060@100 (OC) should give ~150. We know AC68080 is faster per clock (from both Pentium and 060) but it has lower clock than Pentium MMX. GPU with programmable shaders became only popular around ~1200MHz Pentium III and Athlon K7 so there is absolutely no chance to port such games to AC68080.

Again, there are other uses than just porting games...

[Seiya]

Originally Posted by meynaf

Again, there are other uses than just porting games...

Gunnar said that he loves games, so vampire development is very much based on this aspect.
Amiga today is used only for: games.
The future of Amiga is: games

Amiga is retro gaming system and if there are so many beautiful applications, seriously things to do, for 99,9% it is used only for games.