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
Performance and Benchmark Results!

VAMPIRE = the Fastest AMIGA GFX Cardpage  1 2 

Gunnar von Boehn
(Apollo Team Member)
Posts 6207
17 Sep 2017 23:47


Here are latest P96Speed GFX performance scores of the VAMPIRE2.

As you see VAMPIRE provides the fastest AMIGA GFX Card performance.




M Rickan

Posts 177
18 Sep 2017 02:43


Nicely done.

I'm curious... these results are great for reimplementing the original Amiga graphics subsystem.

Would you continue to internalize this functionality within Apollo or eventually favour moving to a dedicated GPU for next generation capabilities?


Steve Ferrell

Posts 424
18 Sep 2017 03:36


m rickan wrote:

    Nicely done.
   
    I'm curious... these results are great for reimplementing the original Amiga graphics subsystem.
   
    Would you continue to internalize this functionality within Apollo or eventually favour moving to a dedicated GPU for next generation capabilities?
   

   
I'd be happy to see a GPU implementation of either type.  It's probably easier (and cheaper) to just use a larger FPGA and enhance/add GPU capabilities into the existing core, so my first guess is that we'll see an FPGA GPU earlier than a dedicated expansion slot and all the supporting hardware needed for a dedicated GPU card. It's also extremely unlikely that the Vampire core (not to mention the OS) could be adapted to work properly with modern nVidia or AMD GPU's and finding GPU's that ARE supported by classic Amigas such as 3DFX cards is extremely difficult and costly and a modern FPGA GPU implementation would run rings around any of that 20+ year old 3DFX tech anyway.
 
I may be wrong, but I think it'll be a long time, if ever, before we see any Vampires using dedicated GPU hardware.

Legal issues aside, how cool would it be to see the Glide API or a 3DFX chipset implemented in the Vampire's FPGA core?
   


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
18 Sep 2017 06:32


m rickan wrote:

Would you continue to internalize this functionality within Apollo or eventually favour moving to a dedicated GPU for next generation capabilities?

 
This would risk to lower the results.
   
The key strength of the VAMPIRE is that CPU is excellent connected to the memory and that the CPU (68080) has very powerful AMMX GFX instructions.
Using AMMX the CPU can directly do high performance GFX rendering.

Take the time and compare the scores with other AMIGA GFX cards.
Compare them with CYBERVISION64, PICASSO 4, compare them with MEDIATOR results with VOODOO and RADEON.
You will see the VAMPIRE beat all GFX cards used on classics.


Steve Ferrell

Posts 424
18 Sep 2017 06:46


Gunnar von Boehn wrote:

m rickan wrote:

  Would you continue to internalize this functionality within Apollo or eventually favour moving to a dedicated GPU for next generation capabilities?
 

 
  Well, this would risk to lower the results.
 
 
  The key strength of the VAMPIRE is that CPU is excellent connected to the memory and that the CPU (68080) has very powerful AMMX GFX instructions.
  Using AMMX the CPU can directly do high performance GFX rendering.
 
 

Any chance of something like the Glide API or an OpenGL subset being added to the core? I think Alain Thellier hinted that it was possible and swapped some assembly code back and forth with you.



Gunnar von Boehn
(Apollo Team Member)
Posts 6207
18 Sep 2017 06:56


Steve Ferrell wrote:

  Any chance of something like the Glide API or an OpenGL subset being added to the core? I think Alain Thellier hinted that it was possible and swapped some assembly code back and forth with you.
 

 
Lets first clarify what we talk in this benchmark about.
 
P96 Speed is a 2D benchmark.
Picaso96 provides for GFX cards, the 2D GFX operations which are used by AMIGA OS.
 
We talk here about drawing or moving Windows, or drawing operations like e.g for rendering Icons.
And some of the functions indicate how good System to GFX memory rates are, showing you a good indication how fast rendering on MAC Emualtor, of MAME games, or game emulator like SCUMVM, or DOOM can be.
 


Steve Ferrell

Posts 424
18 Sep 2017 08:35


Gunnar von Boehn wrote:

Steve Ferrell wrote:

    Any chance of something like the Glide API or an OpenGL subset being added to the core? I think Alain Thellier hinted that it was possible and swapped some assembly code back and forth with you.
   

   
  Lets first clarify what we talk in this benchmark about.
   
  P96 Speed is a 2D benchmark.
  Picaso96 provides for GFX cards, the 2D GFX operations which are used by AMIGA OS.
   
  We talk here about drawing or moving Windows, or drawing operations like e.g for rendering Icons.
  And some of the functions indicate how good System to GFX memory rates are, showing you a good indication how fast rendering on MAC Emualtor, of MAME games, or game emulator like SCUMVM, or DOOM can be.
 

Oh, I understand that completely.  I'm just looking to the future!



Renaud Schweingruber
(Apollo Team Member)
Posts 378
18 Sep 2017 08:56


That's a great speedup over last public core and driver releases !

Kudos to Henryk and Flype for the driver work and of course to Gunnar for the core in-work.


Anthony Jacques

Posts 11
18 Sep 2017 12:02


Gunnar von Boehn wrote:

And some of the functions indicate how good System to GFX memory rates are, showing you a good indication how fast rendering on MAC Emualtor, of MAME games, or game emulator like SCUMVM, or DOOM can be.

I'm certain you know, but for other readers, the rendering portion of Doom etc. that you're referring to is simply transferring the final rendered scene onto the screen. Due to planar-modes requiring C2P this takes up more time than is expected on other architectures, but this is still far from the majority of the processing in rendering the scene in Doom.

These AMMX improvements in the driver layer likely wouldn't significantly speed up the game rendering of Doom unless someone were to write an AMMX optimised port of Doom.


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
18 Sep 2017 12:22


Anthony Jacques wrote:

I'm certain you know, but for other readers, the rendering portion of Doom etc. that you're referring to is simply transferring the final rendered scene onto the screen. Due to planar-modes requiring C2P this takes up more time than is expected on other architectures, but this is still far from the majority of the processing in rendering the scene in Doom.
 
  These AMMX improvements in the driver layer likely wouldn't significantly speed up the game rendering of Doom unless someone were to write an AMMX optimised port of Doom.

Actually the AMIGA DOOM game does support using CHUNKY RTG screens.
There DOOM uses P96 functions like
- WRITECHUNKYPIXEL()

So having a Fast-AMMX accelerated Driver will also speed up DOOM. :-)

Also there are also C2P functions in the driver - and AMMX provides hardware AMMX accelertion for those too.
Doing C2P with AMMX is 100 times faster than the routines you refer to. ;)


Anthony Jacques

Posts 11
18 Sep 2017 14:01


Gunnar von Boehn wrote:

  So having a Fast-AMMX accelerated Driver will also speed up DOOM. :-)
 

  Yes, but as I said, that isn't where the bulk of the time rendering Doom is:
 
Anthony Jacques wrote:

  Due to planar-modes requiring C2P this takes up more time than is expected on other architectures, but this is still far from the majority of the processing in rendering the scene in Doom.
 

 
Gunnar von Boehn wrote:
Doing C2P with AMMX is 100 times faster than the routines you refer to. ;)
 

  No, I'm refering to all of the other CPU/memory bound operations inside the Doom code that are unrelated to C2P and blitting the final image to screen which aren't in any way affected by these changes, and are why you're not going to see a 100x increase in Doom framerates from these changes.
 
  I know that your initial post was talking very specifically about the system->display memory bandwidth, which is completely correct. I'm just highlighting that is all you're referring to, and not Doom in general getting this order of magnitude faster.


Vojin Vidanovic

Posts 770
18 Sep 2017 15:49


Steve Ferrell wrote:
 
      Any chance of something like the Glide API or an OpenGL subset being added to the core? I think Alain Thellier hinted that it was possible and swapped some assembly code back and forth with you.
   

   
    Exactly. Some combo of soft emu (optimized Wazp3d) and hardcoded instructions bringing decent SAGA Warp3D EMU, or even sticking complete Voodo core to V4 Vamp bringing Warp3D and Glide port to os 3.x or AROS, would be a great way ahead. Its not modrn killer, but is enough 3D and may bring many Vodoo class games which arent bad at all + option to hardware accelerate 3D effects within games, OS or hardware assist rendering/fractal generation/landscape generation/morph.
 
  Last known tested Vodoo 5 6000 4 chip config with SDRAM was able to do Doom 3. Not fastest, but nice looking.It was a rescue monster of old school of 3D, up to its limits, but even couple slower and cheap 3D implementations working in parallel, could resolve the 3D side and be integrated/licensed for some future Vampire, or V4 core, if there is enough space.
 
  SAGA (if include even the CPU) is the complete multimedia solution, that lacks 3D. That would be a good edge to Pi and similar, other FPGA systems, new standard even in AmigaOS world. Last time 9250 brought us joy, and even licensing Radeon 9000 Pro/9100, S3 Trio or Renderion Verite core could be the fast solution. Since Vamps dont have PCI or bigger expansion port, FPGA is the only solution. Even CL-GD546X if Cirrus Logic is in question would be a nice add on and respect to 2D/3D pioneers in real "clone wars". Those who died for 3D for the people.
 


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
18 Sep 2017 16:32


Vojin Vidanovic wrote:

Some combo of soft emu (optimized Wazp3d) and hardcoded instructions bringing decent SAGA Warp3D EMU,

I think, with some love and Vampire4 with HARD-FPU
having some 1997 like 3D performance can be a realistic goal.

I personally also look forward to see a good NEO-GEO emulator on VAMPIRE.
I think running ALL NEO_GEO games 100% fluently on VAMPIRE is a realistic target.




Vojin Vidanovic

Posts 770
18 Sep 2017 17:27


Gunnar von Boehn wrote:

  I think, with some love and Vampire4 with HARD-FPU
  having some 1997 like 3D performance can be a realistic goal.

That will be sheer 080 power with FPU + AMMX2 and SAGA to assist.
Have no doubt about that. But then Voodo is the next logical evolutionary step :-)
 
 
Gunnar von Boehn wrote:

  I personally also look forward to see a good NEO-GEO emulator on VAMPIRE.
  I think running ALL NEO_GEO games 100% fluently on VAMPIRE is a realistic target

  I see these are *ALL* SNK games which bring Arcade level gaming (finally!) which is real nice. And deserve a separate news in gaming section.

But complete graphic solution of today should have some form of 3D, not because of snobinnes. Running higher quality games and rendering faster on same machines othewize, then in software render modes.

Also faster 3D then Virges, S3 Trios and Voodo 3 could make SAGA+3D chip - Vampire overall = the Fastest AMIGA GFX Card in full sense of words.



Steve Ferrell

Posts 424
18 Sep 2017 18:50


Vojin Vidanovic wrote:

Steve Ferrell wrote:
 
      Any chance of something like the Glide API or an OpenGL subset being added to the core? I think Alain Thellier hinted that it was possible and swapped some assembly code back and forth with you.
     

     
      Exactly. Some combo of soft emu (optimized Wazp3d) and hardcoded instructions bringing decent SAGA Warp3D EMU, or even sticking complete Voodo core to V4 Vamp bringing Warp3D and Glide port to os 3.x or AROS, would be a great way ahead. Its not modrn killer, but is enough 3D and may bring many Vodoo class games which arent bad at all + option to hardware accelerate 3D effects within games, OS or hardware assist rendering/fractal generation/landscape generation/morph.
   
    Last known tested Vodoo 5 6000 4 chip config with SDRAM was able to do Doom 3. Not fastest, but nice looking.It was a rescue monster of old school of 3D, up to its limits, but even couple slower and cheap 3D implementations working in parallel, could resolve the 3D side and be integrated/licensed for some future Vampire, or V4 core, if there is enough space.
   
    SAGA (if include even the CPU) is the complete multimedia solution, that lacks 3D. That would be a good edge to Pi and similar, other FPGA systems, new standard even in AmigaOS world. Last time 9250 brought us joy, and even licensing Radeon 9000 Pro/9100, S3 Trio or Renderion Verite core could be the fast solution. Since Vamps dont have PCI or bigger expansion port, FPGA is the only solution. Even CL-GD546X if Cirrus Logic is in question would be a nice add on and respect to 2D/3D pioneers in real "clone wars". Those who died for 3D for the people.
   

Yes, that's exactly what I was getting at.  The 3DFX Glide API was open sourced so there would be no legal issues or costs associated with incorporating it into the Vampire core.



Mallagan Bellator

Posts 393
18 Sep 2017 21:29


Can it display 1920x1080 in a decent refresh rate now?


Niclas A
(Apollo Team Member)
Posts 219
18 Sep 2017 21:40


Its hard to find a lot of scores on 8bit. Can you do a 640x480 16bit too? then i have lots of systems to compare too.

In my spreadsheet only in one test the CV3D beats your scores. But that is in 16 bit. Draw() is 14,715. The rest you beat ;)

But the Voodoo 3 destroys the scores. But that is not technically an Amiga GFX card so you are in the clear :D


M Rickan

Posts 177
19 Sep 2017 03:35


Gunnar von Boehn wrote:

This would risk to lower the results...

Take the time and compare the scores with other AMIGA GFX cards.

Compare them with CYBERVISION64, PICASSO 4, compare them with MEDIATOR results with VOODOO and RADEON.

You will see the VAMPIRE beat all GFX cards used on classics.

Without question, the Vampire will give you the fastest classic Amiga.

But I'm talking about the next generation - not replacing but extending these capabilities by taking advantage of the embedded CPUs and GPUs that are now becoming part of FPGAs.

Do you see this as part of the roadmap for the future or is it beyond what you envisioned for Apollo?




Gunnar von Boehn
(Apollo Team Member)
Posts 6207
19 Sep 2017 07:18


Niclas A wrote:

But the Voodoo 3 destroys the scores. But that is not technically an Amiga GFX card so you are in the clear :D

 
In the AMIGA FUTURE is a test of the VOODOO3 combined with MEDIATOR and Amiga 1200, Blizzard 1260/64 Mhz,
EXTERNAL LINK 
According to AMIGA FUTURE, the VAMPIRE wins most of the tests.


Wawa T

Posts 695
19 Sep 2017 11:05


no wonder. mediator, atleast my 4kd1 provides limited transfer speed over zorro. its good for huge albeit rather static desktiop or some 3d, but thats it.

posts 40page  1 2