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!

World Record ATARI FreeMiNT Kronos 68080-FPUpage  1 2 3 4 5 6 

Gunnar von Boehn
(Apollo Team Member)
Posts 6207
16 Dec 2017 19:35


Thanks Markus,
 
 
Markus (mfro) wrote:

 

  00000232 <.L8>:
    232:  3018            movew %a0@+,%d0
    234:  34c0            movew %d0,%a2@+
    236:  32c0            movew %d0,%a1@+
    238:  b288            cmpl %a0,%d1
    23a:  66f6            bnes 232 <.L8>
 

 

 
OK this is easy.

 
Very simple change and twice as fast would be this:
 


  <L8>:
    move.w (a0)+,d0
    move.w d0,(a2)+
    dbra  D7,L8
 

  D7 as dummy for X count
 
 
Up to four times faster would be this.

  <L8>:
    move.l (a0)+,d0
    move.l d0,(a2)+
    dbra  D7,L8

 
 


Stefan Niestegge

Posts 33
17 Dec 2017 01:54


Further progress has been made. Now the most recent Gold 2.7 beta core enables Kronos to complete the 3D OpenGL test as well.

- x13 core on Amiga 500
- ST RAM test fails because that A500 only has 512k (STRAM Test needs 1M)

Here is the .abh file of the Test in the screenshot
EXTERNAL LINK 

full size:
EXTERNAL LINK 

Thanks to Vincent Riviere for the debug-EmuTOS-Floppy and
to BigGun for adding the missing opcode :thumpsup:




Steve Ferrell

Posts 424
17 Dec 2017 02:23


Wow, 175.9 is impressive!  I wish OS3 had OpenGL support.  Does anyone know if OS3 has any OpenGL support such as miniGL or mesa support?  I've been out of the classic Amiga scene for so long that I've lost touch with such things.


Gregthe Canuck

Posts 274
17 Dec 2017 02:32


Looks good!

Missing opcode??


Wawa T

Posts 695
17 Dec 2017 05:21


Does anyone know if OS3 has any OpenGL support such as miniGL or mesa support?

os3 has storm mesa, minigl and even some tinygl implementation. look on aminet for it.


Steve Ferrell

Posts 424
17 Dec 2017 06:34


wawa t wrote:

Does anyone know if OS3 has any OpenGL support such as miniGL or mesa support?

  os3 has storm mesa, minigl and even some tinygl implementation. look on aminet for it.

Oh, nice...maybe a Vampire owner could run some storm mesa benchmarks for us.  I'd be interested in the fps number from the gears demo.


Vojin Vidanovic

Posts 770
17 Dec 2017 08:39


Steve Ferrell wrote:

    Oh, nice...maybe a Vampire owner could run some storm mesa benchmarks for us.  I'd be interested in the fps number from the gears demo.
 

 
  StorMESA 3.0 could work as it supports AGA and GFX cards.
 
  I am not sure about TinyGL requirements, but essentially without SAGA being speeded up it would require a 3D chip to do things hardware based and fast, not CPU rendered. Or in other words, even if we could run it, results could be quite low (as compared to Amiga with Voodo3 card, compared to plain 040+AGA it should be way faster).

Note: if any of them is Warp3d depended, we only have software based WarpZ3d that makes it "worK" (as being run) but also bugs performance down the road.

Beta testers have shown improved 080 optimized WarpZ3d that uses V2 mixed FPU and has decent results, but there is no public release of it (yet).

In short, dont expect 3D performance out of current Vamp line due to mix of hardware probs (no 3D hardware in FPGA) and software (unoptimised and old Amiga implementations requiring mostly AGA or Warp3D dependent).
 


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
17 Dec 2017 09:57


Vojin Vidanovic wrote:

In short, dont expect 3D performance

 
Please don't hijack this topic.
If you want to talk about 3D performance than go to the "TEXTURE MAPPING THREAD". There you can see videos of 3D performance on the VAMP.
 
But please respect the topic of this thread here.
This threads is about ATARI OS  / EMUTOS.
I think is very impressive how nice EMUTOS looks and runs.
 
 
 


Vojin Vidanovic

Posts 770
17 Dec 2017 10:07


Gunnar von Boehn wrote:

  But please respect the topic of this thread here.
  This threads is about ATARI OS  / EMUTOS.
  I think is very impressive how nice EMUTOS looks and runs. 
 

 
  Will go no further - while yes, I admit that AMMX(2) speeds some things up, its only partial 3D hardware assistance (With being no 3D expert I believe effects and shadings are even more consuming then textures). Yes, its really nice to have "fastest Atari too". Any plans to try to go native with MacOS Classic 68k or you are satisfied with Amiga emulation?
 


Markus (mfro)

Posts 99
17 Dec 2017 10:24


Gunnar von Boehn wrote:

... 

    <L8>:
      move.w (a0)+,d0
      move.w d0,(a2)+
      dbra  D7,L8
 

...

and perhaps


    move.w (a0)+,(a2)+

plus loop unrolling would be even faster. Then - as you said - you could also do LONG copies with proper alignment (the good old Duff's device) or even use the Vampire's blitter instead.

As Vincent said, this is based on a (very simple) fVDI example that wasn't optimized for speed, but as a tutorial example for driver writers.

I guess it's not too bad considering its proof of concept state.


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
17 Dec 2017 11:27


Markus (mfro) wrote:

 
Gunnar von Boehn wrote:

    ... 

      <L8>:
        move.w (a0)+,d0
        move.w d0,(a2)+
        dbra  D7,L8
     

    ...

   
    and perhaps
   

    move.w (a0)+,(a2)+
 

  plus loop unrolling would be even faster.
 

Yes, the unrolling would help an 68040 CPU.
But 68080 can execute DBRA for free.
They way we proposed it, the DBRA is executed for free.
This means unrolling is not needed, as faster than FREE we will not get it. :-)
 
Markus (mfro) wrote:

Then - as you said - you could also do LONG copies with proper alignment
 

Yes LONG access would speed up more.
But we don't even not need to worry about the alignment.
APOLLO supports unaligned memory access, and UNALIGNED READS are even free.
 
 
I agree that the current VDI driver is a great start.
I would propose to write the inner DRAW functions in ASM,
or convert the C to ASM with the compiler, and then hand tune the most inner loop.
 
Handtuning FILL and COPY - will only need to change 5-10 instructions. With minimal effort we should get 50% faster.
I look forward to this and think then ATARI will fly.
 


OneSTone O2o

Posts 159
18 Dec 2017 17:10


Atari will fly if a Vampire card plugged in a real ST will boot whatever by the underlying mainboard supported TOS is, and emutos, mint and so on. The most of the Atari scene unfortunatelly (but otherside understandable) will ignore all the phantastic results here as long the underlying mainboard is an Amiga. A year ago this was announced, but, yeah, I know, also a year ago A1200 version was announced, also still not available. So still a long way to go until it really flys. :)


Vojin Vidanovic

Posts 770
18 Dec 2017 17:56


oneSTone o2o wrote:

  The most of the Atari scene unfortunatelly (but otherside understandable) will ignore all the phantastic results here as long the underlying mainboard is an Amiga. A year ago this was announced,

I suppose A500 version and some special core that will utilize ports and other Atari h/w, so I suppose once Amiga v4 standalones get some mature core release, Atari gets to the list :-) It would be interesting market to modernize Ataris :-)


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
18 Dec 2017 21:32


oneSTone o2o wrote:

  A year ago this was announced, but, yeah,
 

   
  I think you misunderstood this.
   
  Yes, the VAMP can physically be put into an ATARI.
  This is clear.
   
  Our testing team are all experienced in AMIGA and we are busy testing several new cards for AMIGA.
  Unfortunately none of us has good ATARI skills, therefore testing on ATARI can without experience not be a priority.
  If experienced ATARI fans step up and help test then VAMP can be made usable for ATARI in short time.
 
 
 
oneSTone o2o wrote:

  I know, also a year ago A1200 version was announced, also still not available.
 

 
  As you certainly know an A1200 card needs the special A1200 CPU connector.
  These connectors are not sold anymore.
  Therefore we had not only to design the V1200 but also had solve the connector issue and had to get molds produced to be able to manufacture our own A1200 connectors. As you see we had a lot to do.


Vojin Vidanovic

Posts 770
18 Dec 2017 22:25


Gunnar von Boehn wrote:

  Unfortunately none of us has good ATARI skills, therefore testing on ATARI can without experience not be a priority.
  If experienced ATARI fans step up and help test then VAMP can be made usable for ATARI in short time.

So generally some call for few AtariST/TT/Falconists and extension of the beta testers by most skilled ones could do the magic in, let say, few months?

That is a great news and possible Vamp secondary market.

Hope OneStone could spread this wonderfull possibility news to Atari cult places.


Olivier Landemarre

Posts 147
18 Dec 2017 22:37


Thanks Stefan,
  High improvement!
  I checked your results, small individual tests are really high, Kronos opengl test is still a bit slower for me than CT60 but it is under partial emulation on V2 so very good I think(here abh result with Radeon EXTERNAL LINK )
 
  I hope be able to buy a V4 card to work a bit on it soon!
 
 
 
Stefan Niestegge wrote:

  Further progress has been made. Now the most recent Gold 2.7 beta core enables Kronos to complete the 3D OpenGL test as well.
 
  - x13 core on Amiga 500
  - ST RAM test fails because that A500 only has 512k (STRAM Test needs 1M)
 
  Here is the .abh file of the Test in the screenshot
  EXTERNAL LINK   
 
 
  full size:
  EXTERNAL LINK   
 
  Thanks to Vincent Riviere for the debug-EmuTOS-Floppy and
  to BigGun for adding the missing opcode :thumpsup:
 
 
 

 


Peter Slegg

Posts 22
22 Dec 2017 12:04



Milan060 @50MHz:
Mothercard Perf.: 501
BogoMIPS: 47.77

Vampire:
Mothercard Perf.: 963
BogoMIPS: 91.7

So getting close to x2 of the Milan, it can do 60MHz but scsi stops working for my scanner.


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
22 Dec 2017 12:42


Peter Slegg wrote:

 
  Milan060 @50MHz:
  Mothercard Perf.: 501
  BogoMIPS: 47.77
   
   
  Vampire:
  Mothercard Perf.: 963
  BogoMIPS: 91.7
   
  So getting close to x2 of the Milan, it can do 60MHz but scsi stops working for my scanner.

"Bogo" means literally "nonsense".
Bogomips might be nonesense, a useless value too?
 
I would look at other values to compare.
I think Olivier said that the OPENGL number is a good indicator of reallife speed.




Peter Slegg

Posts 22
22 Dec 2017 13:34


I assumed that but was told a long time ago that it was is to set various things in the Milan (or Mint).


Olivier Landemarre

Posts 147
23 Dec 2017 10:02


Gunnar von Boehn wrote:

Peter Slegg wrote:

   
  Milan060 @50MHz:
  Mothercard Perf.: 501
  BogoMIPS: 47.77
   
   
  Vampire:
  Mothercard Perf.: 963
  BogoMIPS: 91.7
   
  So getting close to x2 of the Milan, it can do 60MHz but scsi stops working for my scanner.
 

  "Bogo" means literally "nonsense".
  Bogomips might be nonesense, a useless value too?
   
  I would look at other values to compare.
  I think Olivier said that the OPENGL number is a good indicator of reallife speed.
 
 

I fully agree Bogomips is nonesense, and opengl number give good indicator in reallife, really nice result.


posts 109page  1 2 3 4 5 6