Overview Features Instructions Performance Forum Downloads Products OrderV4 Reseller Contact

Welcome to the Apollo Forum

This forum is for people interested in the APOLLO CPU.
Please read the forum usage manual.
VISIT APOLLO IRC CHANNEL



All TopicsNewsPerformanceGamesDemosApolloVampireAROSWorkbenchATARIReleases
Performance and Benchmark Results!

OpenGL On Vampire Cardspage  1 2 3 4 5 6 

Andy Hearn

Posts 269
01 Feb 2019 12:54


indeed. good choice of wording :) looking forward to the next as ever :D


Manuel Jesus

Posts 138
31 Jan 2020 17:02


Gunnar said:

"PS1 quality is not enough for my taste.
We should aim for PS2 qualitylevel."

Music to my ears!


A1200 Coder

Posts 54
31 Jan 2020 18:19


Manuel Jesus wrote:

Gunnar said:
 
  "PS1 quality is not enough for my taste.
  We should aim for PS2 qualitylevel."
 
  Music to my ears!

Well this is not a surprise, the PS1 used affine texturemapping, which distorts the textures pretty badly, especially if done on a triangle basis. The earliest CPU texturemappers on PCs also used affine texturemapping. It wouldn't be acceptable by today's standards, no matter how low end GPU one is planning. The PS2 level is good enough. I don't like 3D stuff that much anyway.



Thellier Alain

Posts 116
31 Jan 2020 19:02


Yes but for the moment 68080 dont seems to have a way to directly do perspective texture mapping
I mean something that will do
  texelRGBA=texture(tex0,u,v,w);



Gunnar von Boehn
(Apollo Team Member)
Posts 4454
01 Feb 2020 16:43


thellier alain wrote:

  Yes but for the moment 68080 dont seems to have a way to directly do perspective texture mapping
  I mean something that will do
  texelRGBA=texture(tex0,u,v,w);
 

Sorry, but what point is this comment?
No CPU on the world has this "directly".
You always have to call a function for this.


Thellier Alain

Posts 116
01 Feb 2020 19:08


We agree : just wanted to say that no CPU has  directly PS2 quality
In fact the real question : is software 68080 3D rendering will be fast enough vs ps2 ?


Gunnar von Boehn
(Apollo Team Member)
Posts 4454
01 Feb 2020 19:37


thellier alain wrote:

We agree : just wanted to say that no CPU has  directly PS2 quality
  In fact the real question : is software 68080 3D rendering will be fast enough vs ps2 ?

Actually I don't think that this was ever said or implied.

The PS2 was a pretty interesting and powerful design

Its a 3 Core design.
1st is a "simple" MIPS core
In addition to this you have 2 Vector CPUs which are tuned for SINGLE FLOAT. (The PS2 can only do SINGLE no DOUBLE.)
Its basically a "small" CELL.

The PS2 is not easy to code for but for its time it was extremely well performing.

The Vampire is NOT a playstation killer and not meant to be.
The Vampire is a true 68K machine.
So programmer friendly and easy to code for is the key focus.
Of course the 68080 is clearly the most powerful 68K ever and its FPU can reach a lot more FLOPS than the 68060.
This means the 68080 can do 3D games better than any 68K machine before.
Developers can use the 68080 to create games better looking that PS1, and visually closer to PS2. But realistically the PS2 is stronger.
 


Vladimir Repcak

Posts 208
06 Feb 2020 13:30


thellier alain wrote:

We agree : just wanted to say that no CPU has  directly PS2 quality
  In fact the real question : is software 68080 3D rendering will be fast enough vs ps2 ?

How could it, possibly ?
PS2 has:
300 MHz main cpu
148 MHz : VP0
148 MHz : VP1
--------------
~600 MHz across 3 processors

It can do ~6,000 MIPS (not 170) and its GPU has a fillrate of ~2.5 gigapixels.




Daniel Sevo

Posts 275
06 Feb 2020 14:13


Vladimir Repcak wrote:

thellier alain wrote:

  We agree : just wanted to say that no CPU has  directly PS2 quality
  In fact the real question : is software 68080 3D rendering will be fast enough vs ps2 ?
 

  How could it, possibly ?
  PS2 has:
  300 MHz main cpu
  148 MHz : VP0
  148 MHz : VP1
  --------------
  ~600 MHz across 3 processors
 
  It can do ~6,000 MIPS (not 170) and its GPU has a fillrate of ~2.5 gigapixels.

 

ye, sounds impossible. Even with the Voodo-esque rasterizer planned for Apollo-core. Of course more AMMX instructions could be added for all sorts of things to speed up various 3d calculations. But none of that would help agaisnt PS2 until of course it was turned into an ASIC so it could runt at 800MHz instead of 80 ;-)



Gunnar von Boehn
(Apollo Team Member)
Posts 4454
06 Feb 2020 18:06


Correct.
PS2 was a beast but really very ugly to program.

The 68K is million times nicer to code.
But we have less horse power.

I think the point is the 68080 has a lot more horse power than any 68K before - this means some pretty nice 3D stuff can now be done. And it can be nicely done and in beauty.
So I would focus on nice gameplay and fun and not on billions of vertices.



Vladimir Repcak

Posts 208
07 Feb 2020 01:33


Gunnar von Boehn wrote:

  But we have less horse power.
 
  I think the point is the 68080 has a lot more horse power than any 68K before - this means some pretty nice 3D stuff can now be done. And it can be nicely done and in beauty.
  So I would focus on nice gameplay and fun and not on billions of vertices.
 

  I don't think it is remotely fair to even attempt to compare PS2 vs Apollo.
 
 
Gunnar von Boehn wrote:

  PS2 was a beast but really very ugly to program.
 
Only if one wanted to write a 3D engine that ran in parallel on all 3 processors.
  If you take the simpler route (majority of earlier games anyway), you still have a 300 MHz MIPS processor that you can treat the same way as 68080 - e.g. just a simple serial execution of all pipeline stages.
 
 
 
Gunnar von Boehn wrote:

  The 68K is million times nicer to code.
Well, have you seen Sony's SDKs ? They give you proprietary compilers that generate a really fast code for VP0-1.
  And you get a basic rasterizer library that is pretty well optimized so you can hit the ground running.



Vladimir Repcak

Posts 208
07 Feb 2020 01:45


There is one use-case where it's however possible to compare the resulting performance of the two distinct machines, though:
  1. C compiler on 300 MHz MIPS
  2. Asm code on 85 MHz Apollo.
 
  Because the C compiler in general will create a really sh*tty code, a hand-written Asm and well-pipelined code on Apollo can get to the same performance level as C on the 300 MHz MIPS.
  But, there's still the PS2's GPU with its insane fillrate...
 


A1200 Coder

Posts 54
07 Feb 2020 09:55


Gunnar von Boehn wrote:

  Correct.
    PS2 was a beast but really very ugly to program.
   
    The 68K is million times nicer to code.
    But we have less horse power.
   
    I think the point is the 68080 has a lot more horse power than any 68K before - this means some pretty nice 3D stuff can now be done. And it can be nicely done and in beauty.
    So I would focus on nice gameplay and fun and not on billions of vertices.
   
 

  So much for the goodness of Playstation 2, any design based on a MIPS R5000 CPU can't be good. Although I have always regarded PS2 games to be of exceptional quality, it's now clear to me that the architecture is no good for the coder. Will be easy to surpass it, perhaps not in performance, but otherwise yeah. It also has some crippled FPU.


Gunnar von Boehn
(Apollo Team Member)
Posts 4454
07 Feb 2020 12:25


A1200 coder wrote:

So much for the goodness of Playstation 2, any design based on a MIPS R5000 CPU can't be good. Although I have always regarded PS2 games to be of exceptional quality, it's now clear to me that the architecture is no good for the coder. Will be easy to surpass it, perhaps not in performance, but otherwise yeah. It also has some crippled FPU.

You are right.
The MIPS core is rather poor compared to the APOLLO 68080.
APOLLO is much more advanced with several integer instructions per cycle, dynamic branch prediction and SIMD AMMX.

The MIPS is real ugly to program and has with 300MHz about the same integer performance as the VAMPIRE card.
So here both the V2 and V4 can easily compete with the Playstation 2.

Where the Playstation clearly has the lead is the peak speed of the Vector-Cores.




Salteadorneo Salteador

Posts 20
07 Feb 2020 13:39


I have worked with Mips at the assembler level and it has its stuff, but you can get more than interesting things and you can work with them at that level without much trouble. Its instruction set is clear enough.
 
  Gunnar von Boehn
  Where the Playstation clearly has the lead is the peak speed of the Vector-Cores.
 
  Could vector cores be added to 68080? I propose the name of Hakaze for this cores. :-D


Vladimir Repcak

Posts 208
07 Feb 2020 14:53


What is so bad about MIPS ? Could it be possibly worse than Jaguar's buggy RISCs ?


Gunnar von Boehn
(Apollo Team Member)
Posts 4454
08 Feb 2020 07:21


Vladimir Repcak wrote:

What is so bad about MIPS ? Could it be possibly worse than Jaguar's buggy RISCs ?

About MIPS everything is bad.
Even the name. :-)
MIPS means machine without interlocking pipeline stages.
What does this name mean?
It means the CPU does NOT track Register usage and Dependancies.
This means that the programmer had to do this - or the CPU would simply calc wrong.
MIPS is the absolute design opposite to 68k.

68K has elegant and powerfull ASM instructions.
With one 68K instruction a lot could be "done".
MIPS instructions as RISC chip are of course "weaker"
but the bigger problem is that you as Programmer need to "count"
the dependencies when coding ..

Example:
MUL #5,D0
Add.l D0,D1

On 68K the MUL might take 2 cycles and the ADD will wait until the result is ready.
The 68K does this tracking also over several integer pipelines.
So even if you have 2 Integer pipelines that ADD will not just be put in the 2nd pipelines and run in parallel to the MUL - as it depends on the result.

MIPS "clever" idea was to avoid all this tracking logic in HW and let the coder worry about it.
This makes MIPS rather unfriendly to code.


Vladimir Repcak

Posts 208
08 Feb 2020 07:46


OMG, that is just plain horrible. That is, indeed, actually much worse than Jaguar's buggy RISC chips...

posts 118page  1 2 3 4 5 6