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!

Some Test With LightWave Vampire CPU Vs 040 Fpupage  1 2 

Enrique Martos

Posts 27
24 Jul 2017 15:59


Some time before i'm read here that fpu code for raytracing apps is not faster simply it's more precise to do ray tracing , some days before i'd make some test with lightwave in my V600 v2 vs my A4000 040/40 .
 
  In vampire i'm use int version of Lightwave and in my Amiga 4000 i'm using lightwave fpu version.
 
  The results are that my A4000 fpu is faster than Vampire doing rendering.
 
  Until now my A600 with vampire it's more faster in all task and apps , except when i can use fpu versions in my Amiga4000.
 
  In the other side Scenery Animator Vampire Int versión is faster than A4000 fpu versión, i'm suppose this reason is because Scenery animator not make an intesive use of procedural or real raytracing algoritms .

Doing the same test with Real3d fpu 040 is fast than Vampire 600 int Real3D.

With this post i don't want say .... it's good , it's bad , i'm know that fpu not implemented for different reasons and it's ok for me, i'm only want to show our tests.
 
Here video where i'd make tests, start at 2h .
 
  EXTERNAL LINK


Simo Koivukoski
(Apollo Team Member)
Posts 601
24 Jul 2017 16:55


Enrique Martos wrote:

Some time before i'm read here that fpu code for raytracing apps is not faster

This claim is not true, no one has said so. Talk has been only of Cinema 4D v4.2 (CU Amiga Super CD-ROM #27) with this test:
 
EXTERNAL LINK 

1m40s: INT
1m42s: FPU1 (Phase 5 060 boards)
1m44s: FPU2 (Apollo, Draco)


Mr Niding

Posts 459
24 Jul 2017 16:59


@Enrique Martos

Thanks for sharing your test :)


Enrique Martos

Posts 27
24 Jul 2017 17:06


I'm suppose it's depend of how the render engine make intensive use of fpu instructions and procedural textures. I'm no have experience with Cinema 4D on Amiga.


Enrique Martos

Posts 27
24 Jul 2017 17:08


Thanks, i want make some test with Imagine 3D too.

In lightwave or similar if you use big scenes with big overload over HD during rendering , vampire is faster moving memory blocks for textures etc..


Simo Koivukoski
(Apollo Team Member)
Posts 601
24 Jul 2017 17:09


It would be interesting to see how Vampire performs to 68882@50MHz on LightWave using this same scene:
 
EXTERNAL LINK   
 
68882@50MHz rendering time: 11h 22m 24s using FPU build.
 


Enrique Martos

Posts 27
24 Jul 2017 17:12


Great idea to test, i'll do it and show here results. A 040 at 25 Mhz are twice fast than a 882 50 Mhz.
 
This reference test was made on Lightwave 3.5 , it's little faster than Lightwave 4.0 than i'm use.
 
  If someone can make the same test with an Amiga 060 will be interesting.


Enrique Martos

Posts 27
24 Jul 2017 17:23


At the end of my video link from the first post , in the positon of 4h, you can watch some of my old work with Lightwave an Amiga in the 90s. I have backup for all  , when we have time i'll try to recover al scenes from my old backup and try to re-render in my vampire or share :D


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
24 Jul 2017 17:49


Simo Koivukoski wrote:

Enrique Martos wrote:

  Some time before i'm read here that fpu code for raytracing apps is not faster

  This claim is not true, no one has said so. Talk has been only of Cinema 4D v4.2 (CU Amiga Super CD-ROM #27) with this test:
 
  EXTERNAL LINK   
 
  1m40s: INT
  1m42s: FPU1 (Phase 5 060 boards)
  1m44s: FPU2 (Apollo, Draco)

I agree every benchmark can be different.
For comparison I can offer some CINEMA4D scores of unreleased CORE.
Cinema 4D v4.2 (CU Amiga Super CD-ROM #27)
Cinema 020 Integer CPU version.

Glasletter 320pix = 1:17min
Glasletter 640pix = 4:59 min

Colortext 320pix = 0:28 min
Colortext 640pix = 1:36 min

Colortext 640pix Antialias=full, 4x4 = 23:53 min

--
Maybe someone has the same Cinema4D version, and likes to compare an 68030 or 68040 or 68060?


Enrique Martos

Posts 27
24 Jul 2017 18:21


i can do the 030 and 040 test. I'll post it here later.


ExiE CZEX

Posts 48
24 Jul 2017 20:17


Gunnar von Boehn wrote:

  For comparison I can offer some CINEMA4D scores of unreleased CORE.
  Cinema 4D v4.2 (CU Amiga Super CD-ROM #27)
  Cinema 020 Integer CPU version.

Raytracing program is not proper software to compare speed of FPU and nonFPU versions. FPU versions usually work with different level of accurancy...



Gunnar von Boehn
(Apollo Team Member)
Posts 6207
26 Jul 2017 09:30


ExiE CZEX wrote:

Gunnar von Boehn wrote:

  For comparison I can offer some CINEMA4D scores of unreleased CORE.
  Cinema 4D v4.2 (CU Amiga Super CD-ROM #27)
  Cinema 020 Integer CPU version.
 

  Raytracing program is not proper software to compare speed of FPU and nonFPU versions. FPU versions usually work with different level of accurancy...

Please give some proof for this claim.

As a matter of fact,
a float number inside the FPU is internally
nothing else but 2 INTEGER numbers.

So technically a FLOAT is 2 INTEGERS.
You can do exactly the same calculation on 2 INTEGERS inside or outside of the FPU - and you will have the same accuracy.



Chris Dennett

Posts 67
26 Jul 2017 11:54


Here are some interesting posts:

EXTERNAL LINK  EXTERNAL LINK  EXTERNAL LINK


Obetto Sannala

Posts 61
22 Dec 2017 14:41


Just to let you know an answer from the NewTek / LightWave Support Team to my question "Is it legal, to copy and run the latest Amiga-Version of Lightwave?"

Answer:
Hi there -
If you can find an Amiga version - I say use it.


Mallagan Bellator

Posts 393
22 Dec 2017 17:17


Gunnar von Boehn wrote:

  As a matter of fact,
  a float number inside the FPU is internally
  nothing else but 2 INTEGER numbers.
 
  So technically a FLOAT is 2 INTEGERS.
  You can do exactly the same calculation on 2 INTEGERS inside or outside of the FPU - and you will have the same accuracy.
 

Does this also mean that fpus need twice as long to perform certain instructions as a cpu would? Like cpu without fpu inside, of course. Or do the instructions inside the fpu allow for the fpu to process the task just as fast?



Stefano Briccolani

Posts 586
22 Dec 2017 17:27


@Obetto
 
  Can you post the link about newtek's answer to your question?

That could be a way to put lightwave on aminet..


Stefano Briccolani

Posts 586
01 Jan 2018 15:28


@Obetto Sannala
I had a chat with @emufan at EAB forum. If you can give us an evidence of the newtek reply, or if you're still in contact with newtek support and ask about putting lightwave on aminet, will be an awesome present to amiga community. Lightwave is still a great 3d program.
EXTERNAL LINK


Pedro Cotter
(Apollo Team Member)
Posts 308
22 Feb 2018 07:41


Old news, but we broke some world records already!!!

EXTERNAL LINK 
EXTERNAL LINK 
Test validated by the Polish Amiga Forum
EXTERNAL LINK


Djole Djole

Posts 35
22 Feb 2018 13:16


Pedro Cotter wrote:

Old news, but we broke some world records already!!!
 
  EXTERNAL LINK 
  EXTERNAL LINK 
  Test validated by the Polish Amiga Forum
  EXTERNAL LINK 

Nice results! Why LW 3.5 ?



Pedro Cotter
(Apollo Team Member)
Posts 308
23 Feb 2018 06:25


Djole Djole wrote:

Pedro Cotter wrote:

  Old news, but we broke some world records already!!!
 
  EXTERNAL LINK   
  EXTERNAL LINK   
  Test validated by the Polish Amiga Forum
  EXTERNAL LINK 

 
  Nice results! Why LW 3.5 ?
 

Hi,

It's a requirement for the test validation. Works fine with later version too!


posts 35page  1 2