Overview Features Coding ApolloOS Performance Forum Downloads Products Order Contact

Welcome to the Apollo Forum

This forum is for people interested in the APOLLO CPU.
Please read the forum usage manual.
Please visit our Apollo-Discord Server for support.



All TopicsNewsPerformanceGamesDemosApolloVampireAROSWorkbenchATARIReleases
Information about the Apollo CPU and FPU.

page  1 2 3 4 5 6 7 8 9 10 11 

Roman S.

Posts 149
05 May 2017 06:57


Gunnar, the point with FPU is not only speed of selected few applications - but compatibility. Without FPU I had a lot of problems with software on my A1200/030 - I had to look for specially compiled non-FPU versions of libraries - but some programs were still unreliable (gurus).

Problem is, that people writing something intended to be used with fast machines often assumed the FPU would certainly be there (almost all 040 and 060 have it built-in, A3000 has FPU on the motherboard, and adding FPU to 68030 accelerator is often easy and cheap - not to mention UAE users) - they didn't care if some FPU opcodes were present here and there.

You could patch the popular SysInfo, but you can't hunt down and patch everything.


OneSTone O2o

Posts 159
05 May 2017 07:36


You can dramatically increase the base of 68K software using FPU very easy. You also can increase the number of skilled peoples/programmers who could test FPU in Apollo.
 
The point is, you mostly think beige, not ... grey. Think grey!
 
Grey is the color of a lot more FPU usage. Grey is the color of another community waiting for Vampire accelerators. But grey also needs PMMU. Vampire fits mechanically, but nobody has seen it fireing up in grey boxes yet.
 
Grey computers with 68K processor have been designed in Borregas Avenue, Sunnyvale.
 
_/|\_ It's like a dance on a volcano, a dance on Fujijama!


Thierry Atheist

Posts 644
05 May 2017 07:59


They will make an FPU. (Always were going to.)

Now we know that it's going to be unusual.... Unusually POWERHOUSE!

Why are people complaining?


Saladriel Amrael

Posts 166
05 May 2017 08:22


I believe Apollo team know what they are doing. Maybe we will not get 040 compatible FPU today, but if they tell being working on something more innovative for the future I trust them.

I also think they have a sort of priority schedule, and FPU is not on top of that list (And I can see why, with Paula and AGA being wip).

Vampire is not just a faster accelerator board, but something more complex, unique and complete. It is going to add our Amigas new functionalities. Just chill out, relax, read the news and dream ;)


Obetto Sannala

Posts 61
05 May 2017 08:59


@saladriel

+1


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
05 May 2017 09:59


Roman S. wrote:

Gunnar, the point with FPU is not only speed of selected few applications - but compatibility. Without FPU I had a lot of problems with software on my A1200/030 - I had to look for specially compiled non-FPU versions of libraries

Lets be clear honest here.
There are 4 options of solving this:

a) You compile the software new.

b) You use a software SoftFPU.
There was a Software FPU for the MAC. Port it!
Or use a SoftFPU like in Petunia.

c) You spend the time and money for a compatible hardware FPU.
Yes money because the FPU use major part of the FPGA and there for will also limit other development options.

d) You spend the time and money for a compatible hardware FPU - but also go the effort to make it compatible with future FPGA technologie and go the effort to make it that fast that you can enable software which was never been able to do on AMIGA land.

We are clearly in favour of option D)



Ian Parsons

Posts 230
05 May 2017 10:09


Does that mean a larger FPGA than currently used on the Vampire 500/600 will be needed to have both the FPU and the other development options you want?


Andrew Copland

Posts 113
05 May 2017 11:04


Demos, raytracers (POVray), fractals, audio en/decoding.

Yes you've always been able to use a SoftFPU, but they can drop the float performance by another order of magnitude.

So a 68060 FPU is about 1000 times faster than a 68881, does that mean we're looking at floating point operations using SoftFPU being 10000 times slower? The Vampire is fast but it's not 10000 times faster, or even 1000, not even 100 times faster than a 68060's FPU.

You have a mostly working FPU right now?
What do you need to test it?
What would be different between the FPU now and "option D"?

Andy


Saladriel Amrael

Posts 166
05 May 2017 12:22


Gunnar von Boehn wrote:

 
  Lets be clear honest here.
  There are 4 options of solving this:
 
  a) You compile the software new.
 
  b) You use a software SoftFPU.
  There was a Software FPU for the MAC. Port it!
  Or use a SoftFPU like in Petunia.
 
  c) You spend the time and money for a compatible hardware FPU.
  Yes money because the FPU use major part of the FPGA and there for will also limit other development options.
 
  d) You spend the time and money for a compatible hardware FPU - but also go the effort to make it compatible with future FPGA technologie and go the effort to make it that fast that you can enable software which was never been able to do on AMIGA land.
 
  We are clearly in favour of option D)
 

That's very good to me.
This means that:
1)Apollo team is developing an FPU
2)Apollo team is aiming to have a more efficient FPU than Motorola ever managed to put in their chips when 68k line was still "alive"
3)Apollo team is looking at the future in a way that they'll be able to get advantage from new FPGA technology

If this means to wait in order to have a fully developed FPU into Nat... hem Vampire then I'm on board and fully supporting.


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
05 May 2017 12:35


Hi Andrew,
 
 
Andrew Copland wrote:

  The 68060 FPU has some short falls.
 

The 68060 is not fully pipelined.
This means FADD throughput is 1/3 clocks.
 
 
1) Our FPU is FULLY pipelined.
We can have a throughput of 1 FMAD per clock.
 
2) Our FPU units are independent and can work in parallel.
This means we can have a throughput of not only 1 FADD per clock.
But also 1 FMUL per clock.
This means with good code you have 2 FLOPS per cycle.
 
3) The 68K FPU design is limited by 2 operants ISA encoding.
The 2 Operant encoding create the need for many "filler" FMOVE instructions. These instruction create unneeded dependency chains and lower the FPU performance.
 
Our is internally a 3 Operant machine. By for example FUSING FMOVE+FADD or FMOVE+FMUL we can create a much higher throughput than all previous 68K-FPU.
 
If you take the above differences into account the you see that we can provide 10 times higher performance than 68060.
 
But today there is NO software written to use this power.
To use this full power potential, the software needs to be be written use the Super-Scalarity really efficiently - this is best done in ASM.
   
There are also some challenges in writing this.
As while the throughput is super, all FPU instructions still have a latency. This means perfect FPU code needs to consider this.

A good coder good do with something - never seen on AMIGA before.
If coder show the will to do this, to write a small but Killa-app, and has the will to test/profile/debug the FPU design with this - which can take several month. Then we will happily support this.


OneSTone O2o

Posts 159
05 May 2017 13:22


As long as this new Apollo FPU is still fully compatible to existing FPU based software, every thing is fine.


Obetto Sannala

Posts 61
05 May 2017 13:23


I'm sorry - this is probably a silly question. I am not a coder and I have no clue about FPGA development.

But could this be a future idea (?):

Old 68k software needs FPU-> interface <-Apollo FPU


Tango One

Posts 102
05 May 2017 15:35


When I used my amiga everyday from 1985 to 1994 I think it was maybe 1-2 programs that wanted a fpu.

Maybe fpu is something we can think about when all amiga have a vampire.(hint)




Michal Warzecha

Posts 209
05 May 2017 15:49


tango one wrote:

When I used my amiga everyday from 1985 to 1994 I think it was maybe 1-2 programs that wanted a fpu.
 
  Maybe fpu is something we can think about when all amiga have a vampire.(hint)
 
 
 
 

So, very soon I think :)
I'm reading this discussion but I'm only Amiga-gamer to be honest, so, FPU is not necesari for me. But I can see many ppl really want it, even if They can't use it much. If FPU is working, can You share core with FPU to let them play with? Even if it's buggy, but if they want it so much. It's only sugestion. In my opinion to write anything people need hardware to work with :)



Andrew Copland

Posts 113
05 May 2017 16:07


tango one wrote:

When I used my amiga everyday from 1985 to 1994 I think it was maybe 1-2 programs that wanted a fpu.
Maybe fpu is something we can think about when all amiga have a vampire.(hint)

An FPU is just about taking some processing off the integer part of the CPU itself using dedicated hardware (the FPU) to handle floating point maths.

TFX might be another game to benefit from it, possibly Birds of Prey.

The lack of an FPU as standard always created a chicken and egg problem for Amiga. It wasn't there as standard so no-one used it thus you get the "no software uses it" argument.

Gunnar I'm glad to see that what you've got is already quite capable, much like the AMMX stuff though, if you don't release it then it can't be used.


Stefan Niestegge

Posts 33
05 May 2017 16:58


What about Demos? Don't the Amiga Demos need an FPU? I was hoping to be able to watch Amiga AGA 060 demos on vampire upgraded Amigas one day.

At least the "060 only" stuff always had the FPU available on the 060 upgrades (Have there even been cards with fpu-less EC68060 CPUs?




Gunnar von Boehn
(Apollo Team Member)
Posts 6207
05 May 2017 17:48


Andrew Copland wrote:

Gunnar I'm glad to see that what you've got is already quite capable, much like the AMMX stuff though, if you don't release it then it can't be used.

I think my points it not clear to understand.
The old 68k FPU does calc everything in 80bit.

Does anything on AMIGA software needs this? NO!
Does any other software in the world need this? NO!
Other Architectures like IBM machines can not even do this.

No software _really_ needs this.

But our FPUs does provide this.
But of course we pay for this with cost and latency.
I now see in the near future the option to use faster FPGAs.
FPGAs so fast that we can reach GIGAFLOPS with them.
Then games on a complete new level could be possible on AMIGA/APOLLO.
Now these new FPGA will not be able to support the 80bit FPU.

So we are between a rock and hardplace now.
We can put work into releasing an 80bit compatible FPU - which we know that will drop theis 80bit in 1 year.

Or we try to think of a clever solution for this.


Szyk Cech

Posts 191
05 May 2017 18:20


Gunnar von Boehn wrote:

  So we are between a rock and hardplace now.

But what about IEEE standards for the floating point operations?!? It is not good enough for Apollo Core?!? If MMU breaks compatibility (it's no pity - even Moto do this), so maybe FPU can be tuned in that way?!?

But thanks for the info that major FPGA space is required for the FPU - I don't realize this...



Kolbjørn Barmen
(Needs Verification)
Posts 219/ 2
05 May 2017 18:54


Gunnar von Boehn wrote:

  The old 68k FPU does calc everything in 80bit.
 
  Does anything on AMIGA software needs this? NO!

 
  Wrong question - does any Amiga software _use_ this?
 
 

  But our FPUs does provide this.
  But of course we pay for this with cost and latency.
 

 
  No problem - right now compatibility with existing software and existing vampire is much, much more important than...
 
 
I now see in the near future the option to use faster FPGAs.
  FPGAs so fast that we can reach GIGAFLOPS with them.

 
  ... when hopefully the entire "apollo core" thing has matured, in terms of developers and new software.
 
 

  We can put work into releasing an 80bit compatible FPU - which we know that will drop this 80bit in 1 year.
 

 
  Do it.
 
 
 
  Or we try to think of a clever solution for this.
 

 
  Those things are known as "roadmap".
 
  Provide a backwards compatible FPU now, so that those who have current Vampire cards and software that requires FPU, can use them together as soon as possible. Next year, when you have new amazing Vampire cards available, with larger and faster FPGAs, and a new core with new FPU and who knows what... those who want to upgrade can do that, and developers will also know before hand.
 
  And in that time period when some people have "old" (current) Vampire cards that only do "old legacy" FPU, and some people have upgraded to newer faster FPGAs and new FPU, someone may do something clever like a "soft legacy FPU" that makes use of whatever resources that is available in the newer core.
 
  The "wrong" thing to do, is to not provide any FPU to current owners who have read about the FPU and been told about its greatness for three years.


Kolbjørn Barmen
(Needs Verification)
Posts 219/ 2
05 May 2017 19:04


Thierry Atheist wrote:

  Why are people complaining?

Because, chances are that it will not be able to run existing software that requires FPU. Maybe you are someone who only used Amiga for games, but quite a few of us never really used it much for games as we used it for various kinds of arts and production. Like 3D modelling and raytracing, manipulation of graphics and audio. Animation. You know, all that stuff that is also used to _create games_, not just play them.

posts 206page  1 2 3 4 5 6 7 8 9 10 11