3dfx + swShader

[r21vo]

today I was browsing net and searching for 3dfx sites I haven't seen yet and found some project (http://sw-shader.sourceforge.net) about developing shaders in software. Yeah I know - shaders in software - it's realllllly sloooooow, BUT, look at cpus now - thier speed is increasing  more and more. Maybe it's worth to try out swShaders lib?

I was wondering maybe with very fast cpu some shaders could be done in software fast enough.. How do you think?

[secretfj]

what voodoo cards lack is Pixel Shader...
the oldest mainstream gaming card that implement Pixel Shader is GeForce 3...
the problem is, it's still quite powerful compared to nowadays cards and CPUs..
also seems it's still not a must to have PS in nowadays games

for Vertex shader, Voodoo cards does have some VS unit

maybe some programs that force HW T&L games to run at SW T&L (Direct X does provide software T&L support, as you can see in 3D Mark 2001, and it's NOT null rendering) would be nice~~

[r21vo]

Originally posted by secretfj

what voodoo cards lack is Pixel Shader...
the oldest mainstream gaming card that implement Pixel Shader is GeForce 3...
the problem is, it's still quite powerful compared to nowadays cards and CPUs..

is that a problem? And of course it's powerful compared to low-end gayforce cards..

Originally posted by secretfj


also seems it's still not a must to have PS in nowadays games

with 3DAnalyzer it's not a must , but many games use pixel shaders. Some games can be played without them, but it spoils gameplay.. :/

Originally posted by secretfj


maybe some programs that force HW T&L games to run at SW T&L (Direct X does provide software T&L support, as you can see in 3D Mark 2001, and it's NOT null rendering) would be nice~~

You mean developing it in drivers so games can use HW T&L which actually is SW T&L?

[secretfj]

It's mt bad expression..
I mean to make a CPU to do pixelshader (2.0 for DX9C) is maybe a bit too hard and the result maybe be no good

I have to admit that without PS, the gaming experience is deteriorated(lack of special effect like beautiful water waves done by PS in common)

and i haven't considered PS optimized games like farcry..

but compared with software-based PS, T&L is easier and already done by DirectX (i think someone had said it before), it's just the game maker won't use it in case of display cards that lack of T&L unit...so implement a cheat to make the game run but intercept those T&L commands to DirectX built-in software T&L Emulator

P.S. maybe someone will write a translator to translate/optimize those PS/T&L codes into Glide and feed it to VooDoos, it would be nice
(does MESAfx do the similar thing? translating those OpenGL connands into Glide so that Voodoo can use its power to handle it..i dunno)

[r21vo]

Originally posted by secretfj

It's mt bad expression..
I mean to make a CPU to do pixelshader (2.0 for DX9C) is maybe a bit too hard and the result maybe be no good

I thought about software shaders as optional feature which can be enabled/disabled. I guess many of us would like to see how voodoo runs with shaders.

Originally posted by secretfj


I have to admit that without PS, the gaming experience is deteriorated(lack of special effect like beautiful water waves done by PS in common)

for example famous NFS:U - without special effects it's not NFS:U (I played that game on my gf4ti4200 with all effects on, and when i see it without them, it looks awful to me.. :/ )

Originally posted by secretfj


P.S. maybe someone will write a translator to translate/optimize those PS/T&L codes into Glide and feed it to VooDoos, it would be nice

first we need someone who knows how to add swShaders to drivers..

[secretfj]

the creator does admit that it's a pre-alpha one...
but i tried one of the demo..3X fps for a car model...i think it's very good for merely software rendering using PS..

I am looking forward to this because my Radeon 9100VE won't play any PS2.0 stuff (you are lucky, due to my out-dated system, i can't turn all the details on..so have no idea how good i would be..)

[r21vo]

Originally posted by secretfj

the creator does admit that it's a pre-alpha one...
but i tried one of the demo..3X fps for a car model...i think it's very good for merely software rendering using PS..

I am looking forward to this because my Radeon 9100VE won't play any PS2.0 stuff (you are lucky, due to my out-dated system, i can't turn all the details on..so have no idea how good i would be..)

just tested that car model demo (Per-Pixel Lighting from PS2.0) - seems very nice. I got about ~25fps (when moving around the scene) on my AthlonXP 2200+..
in StencilShadow (stencil buffer capabilities) demo i got ~25fps when not moving and ~13fps when flying around.
in ShaderVolume (DX9 SDK) i got about 15fps.

also i tried RealVirtuality demo which illustrates use of SoftWire Run-Time Assembler library - got about ~15fps when moving around.

when looking on these numbers the idea of software shaders seems promising..

[secretfj]

some good news from the SW shader developers:
they released some Direct 3d Dll for demo/testing, and i did some test using normal direct3D tester -- the fillrate benchmark work and some results are gained, but for others like shadermark 2.1 or PS tester, the result is still negative.
but seems there is big hope for software shaders, let's be positive

some result(actually some minor render error occured in the result done by SW Shader)

FillrateBenchmark(tm) 2004 - "easy benchmark series"

   Benchmark Main Program Version: FRB_V092
   Benchmark Date/Time : 1/1/2005 22:57:26

                    System Information
-----------------------------------------------------------
       CPU : VIA Nehemiah
       GFX : swSHader
        OS : Microsoft Windows XP
  Settings : 1024x768  16 bits  D16  No AA

                     Benchmark Result
-----------------------------------------------------------
          FrameBuffer Clear : 102.4 FPS
                 Color Fill : 25.16582 M-Pixel/s
                     Z Fill : 12.58291 M-Pixel/s
             Color + Z Fill : 15.09949 M-Pixel/s
             Single Texture : 7.549747 M-Pixel/s
 Single Texture Alpha Blend : 5.033165 M-Pixel/s
              Dual Textures : 5.033165 M-Pixel/s
            Triple Textures : 5.033165 M-Pixel/s
              Quad Textures : 5.033165 M-Pixel/s
   1 Floating Poing Texture : 7.549747 M-Pixel/s
             Render to Self : 0.8388608 M-Pixel/s
              PS 1.1 Simple : 7.549747 M-Pixel/s
              PS 1.4 Simple : 7.549747 M-Pixel/s
              PS 2.0 Simple : 7.549747 M-Pixel/s
           PS 2.0 PP Simple : 7.549747 M-Pixel/s
    Customized Pixel Shader : 7.549747 M-Pixel/s
             PS 2.0 Complex : (Unsupported)
          PS 2.0 PP Complex : (Unsupported)
    PS 2.0 Massive Register : (Unsupported)
 PS 2.0 PP Massive Register : (Unsupported)
PS 2.0 Sincos Procedure Tex : (Unsupported)
  PS 2.0 Per-Pixel Lighting : (Unsupported)
-----------------------------------------------------------
   * End of FillrateBenchmark Result

Reference result done by Voodoo3 3000

FillrateBenchmark(tm) 2004 - "easy benchmark series"

   Benchmark Main Program Version: FRB_V092
   Benchmark Date/Time : 1/1/2005 23:00:25

                    System Information
-----------------------------------------------------------
       CPU : VIA Nehemiah
       GFX : 3dfx Interactive, Inc. Voodoo3 Series Driver
        OS : Microsoft Windows XP
  Settings : 1024x768  16 bits  D16  No AA

                     Benchmark Result
-----------------------------------------------------------
          FrameBuffer Clear : 780.8 FPS
                 Color Fill : 201.3266 M-Pixel/s
                     Z Fill : 198.81 M-Pixel/s
             Color + Z Fill : 198.81 M-Pixel/s
             Single Texture : 198.81 M-Pixel/s
 Single Texture Alpha Blend : 196.2934 M-Pixel/s
              Dual Textures : 196.2934 M-Pixel/s
            Triple Textures : 196.2934 M-Pixel/s
              Quad Textures : (Unsupported)
   1 Floating Poing Texture : (Unsupported)
             Render to Self : (Unsupported)
              PS 1.1 Simple : (Unsupported)
              PS 1.4 Simple : (Unsupported)
              PS 2.0 Simple : (Unsupported)
           PS 2.0 PP Simple : (Unsupported)
    Customized Pixel Shader : (Unsupported)
             PS 2.0 Complex : (Unsupported)
          PS 2.0 PP Complex : (Unsupported)
    PS 2.0 Massive Register : (Unsupported)
 PS 2.0 PP Massive Register : (Unsupported)
PS 2.0 Sincos Procedure Tex : (Unsupported)
  PS 2.0 Per-Pixel Lighting : (Unsupported)
-----------------------------------------------------------
   * End of FillrateBenchmark Result

software does gives more functionality while old hardware has its own power over software, so maybe combining them together is the best
(makes me think of Dborca and MESAFX, MESA is a pure SW openGL substitute, and Daniel Borca give the software the power of Voodoo cards, maybe we need another genius like Daniel - I am sure there are some in Voodoo players~~)

[r21vo]

just tested my main rig with athlonxp 2500+ and gf4ti4200-8x to compare results.

here are results of my chaintech gf4ti4200-8x 128Mb:

   FillrateBenchmark(tm) 2004 - "easy benchmark series"

   Benchmark Date/Time : 2005.01.02. 01:58:05

                    System Information
-----------------------------------------------------------
      CPU : AMD Athlon(tm) XP 2500+
      GFX : NVIDIA GeForce4 Ti 4200 with AGP8X
       OS : Microsoft Windows XP
 Settings : 1024x768 16 bits D16 No AA

                     Benchmark Result
-----------------------------------------------------------
          FrameBuffer Clear : 2497,6 FPS
                 Color Fill : 999,0833 M-Pixel/s
                     Z Fill : 999,0833 M-Pixel/s
             Color + Z Fill : 999,0833 M-Pixel/s
             Single Texture : 981,4672 M-Texel/s
              Dual Textures : 1889,953 M-Texel/s
            Triple Textures : 1458,045 M-Texel/s
              Quad Textures : 1897,503 M-Texel/s
   1 Floating Poing Texture : (Unsupported)
             Render to Self : 582,7985 M-Pixel/s
              PS 1.1 Simple : 999,0833 M-Pixel/s
              PS 1.4 Simple : (Unsupported)
              PS 2.0 Simple : (Unsupported)
           PS 2.0 PP Simple : (Unsupported)
             PS 2.0 Complex : (Unsupported)
          PS 2.0 PP Complex : (Unsupported)
    PS 2.0 Massive Register : (Unsupported)
 PS 2.0 PP Massive Register : (Unsupported)
PS 2.0 Sincos Procedure Tex : (Unsupported)
  PS 2.0 Per-Pixel Lighting : (Unsupported)
-----------------------------------------------------------
   * End of FillrateBenchmark Result

here are results of swShader:

   FillrateBenchmark(tm) 2004 - "easy benchmark series"

   Benchmark Date/Time : 2005.01.02. 02:02:16

                    System Information
-----------------------------------------------------------
      CPU : AMD Athlon(tm) XP 2500+
      GFX : swSHader
       OS : Microsoft Windows XP
 Settings : 1024x768 16 bits D16 No AA

                     Benchmark Result
-----------------------------------------------------------
          FrameBuffer Clear : 659,4 FPS
                 Color Fill : 138,412 M-Pixel/s
                     Z Fill : 70,46431 M-Pixel/s
             Color + Z Fill : 75,49747 M-Pixel/s
             Single Texture : 5,033165 M-Texel/s
              Dual Textures : 5,033165 M-Texel/s
            Triple Textures : 5,66231 M-Texel/s
              Quad Textures : 7,549747 M-Texel/s
   1 Floating Poing Texture : 5,033165 M-Pixel/s
             Render to Self : 0,4194304 M-Pixel/s
              PS 1.1 Simple : 128,3457 M-Pixel/s
              PS 1.4 Simple : 125,8291 M-Pixel/s
              PS 2.0 Simple : 125,8291 M-Pixel/s
           PS 2.0 PP Simple : 125,8291 M-Pixel/s
             PS 2.0 Complex : (Unsupported)
          PS 2.0 PP Complex : (Unsupported)
    PS 2.0 Massive Register : (Unsupported)
 PS 2.0 PP Massive Register : (Unsupported)
PS 2.0 Sincos Procedure Tex : (Unsupported)
  PS 2.0 Per-Pixel Lighting : (Unsupported)
-----------------------------------------------------------
   * End of FillrateBenchmark Result

also i noticed that results did't changed (and picture too) while changing color depth when i tested swShader. While testing my gfx results and image changed much.

so from the results it's easy to see that even on athlonxp 2500+ shShader is slower than stock voodoo3 running on via c3 900mhz. Although the results of software shaders are better than i expected, maybe combining hardware and software could do the thing (emulating only shaders and some other caps that voodoo doesn't support).

[secretfj]

oh sorry i've forgotten to say, the V3 is overclocked to 179/179MHz

[Nightbird]

I quote here the author :

swShader is a piece of software which allows to render advanced 3D scenes without the need for specialized hardware

While most gamers do have the latest generation of graphics card, this is not the case for the average computer user. However, advanced 3D graphics is spreading fast, and many of these people are left in the cold. Keeping up with the latest graphics upgrade is an uphill battle, or not a battle at all for unaware consumers

Like 3DAnalyzer emulates perfectly some functions, swShader can do probably the same since it uses the cpu...

[r21vo]

Originally posted by secretfj

oh sorry i've forgotten to say, the V3 is overclocked to 179/179MHz

anyway, oc'ing v3 doesn't change results very much

Originally posted by Nightbird

Like 3DAnalyzer emulates perfectly some functions, swShader can do probably the same since it uses the cpu...

difference is that 3DAnalyzer emulates capabilities and puts a null rendering but swShader renders all in software.

written in 3danalyzer docz


Yet, this is presently achieved not by Emulation but by NUL-Rendering.

also i found out that 3danalyzer uses the same technique swShader uses to get between 3d api and gfx - replacing direct3d dll, only 3danalyzer's dll somehow inherits all cap's of gfx + options which you select in soft, but shShader dll doesn't have any connection to gfx. Maybe it's possible to integrate shShader pixel shader rendering source into 3danalyzer - replace null rendering with shShader software rendering?

[Nightbird]

Nobody is interested to program the perfect utility ?

[secretfj]

Originally posted by Nightbird

Nobody is interested to program the perfect utility ?

my own experience tells me that SW shader still have to improve its competability with DX9 programs/games/benchmarks -- majority of them still gives error message when using SWshader. but anyway it's a good start

[Nightbird]

but anyway it's a good start

You're right and sounds very promising.
For the Author, keep the good work !