X1600 MSAA Fillrates

Unknown Soldier · Jan 4, 2006

Cat5.13 running at 520/560, nothing else running in the Background.

Code:

 FillrateBenchmark(tm) 2004 - "easy benchmark series"

    Benchmark Date/Time : 04/01/2006 19:46:42

                     System Information
-----------------------------------------------------------
       CPU : AMD Athlon(tm) XP 3200+
       GFX : RADEON X800 XT Platinum Edition AGP
        OS : Microsoft Windows XP

                      Benchmark Result - Resolution: 1024x768 - 85Hz
-----------------------------------------------------------

No AA
           FrameBuffer Clear : 6596.2 FPS
                  Color Fill : 5981.916 M-Pixel/s
                      Z Fill : 7854.254 M-Pixel/s
              Color + Z Fill : 4547.464 M-Pixel/s
-----------------------------------------------------------

2xAA
           FrameBuffer Clear : 6227.4 FPS
                  Color Fill : 5903.903 M-Pixel/s
                      Z Fill : 6850.138 M-Pixel/s
              Color + Z Fill : 4109.579 M-Pixel/s
-----------------------------------------------------------

4xAA
           FrameBuffer Clear : 5791.2 FPS
                  Color Fill : 5496.215 M-Pixel/s
                      Z Fill : 3925.869 M-Pixel/s
              Color + Z Fill : 3379.77 M-Pixel/s
-----------------------------------------------------------
Non-Maskable AA
           FrameBuffer Clear : 6227.4 FPS
                  Color Fill : 5903.903 M-Pixel/s
                      Z Fill : 6855.17 M-Pixel/s
              Color + Z Fill : 4107.063 M-Pixel/s

PeterAce · Jan 4, 2006

Here is G70 again, but on the BFG GTX OC 512MB (580Mhz Core, 1750 Mhz Memory - stock settings) :

Code:

   CPU : AMD Athlon(tm) 64 Processor 3500+
        GFX : NVIDIA GeForce 7800 GTX
         OS : Microsoft Windows XP

   Settings : 1024x768  32 bits  D16  No AA

                  Color Fill : 8596.646 M-Pixel/s
                      Z Fill : 17178.19 M-Pixel/s
              Color + Z Fill : 6613.578 M-Pixel/s
     
   Settings : 1024x768  32 bits  D16  2x FSAA

                  Color Fill : 8596.646 M-Pixel/s
                      Z Fill : 8596.646 M-Pixel/s
              Color + Z Fill : 6450 M-Pixel/s

   Settings : 1024x768  32 bits  D16  4x FSAA

                  Color Fill : 4572.63 M-Pixel/s
                      Z Fill : 4572.63 M-Pixel/s
              Color + Z Fill : 3447.718 M-Pixel/s

As expected the optimized 'Double Z when no Colour' is disabled with 2xAA or anything higher.

Hmm.. the colour fillrate seams ROP limated (580 x 16 ROPS) or is memory bandwidth the limating factor?

Jawed · Jan 4, 2006

We still don't know if Dave is testing at D16 or D24S8 or, erm... though it isn't making a huge difference to my X800XT.

Peter you might find D24S8 gives you slightly higher performance - it often does with my X800XT (though not comprehensively).

It's noteworthy that the GTX-512's Z-only fillrate is 185% of the theoretical colour fillrate, as compared with a standard GTX at 176%. But the 2xAA and 4xAA results are worse (though only marginally), as are the colour rates.

On a sidenote, I think the X1800XL Z fillrate reported earlier in this thread at 4xAA must be a driver glitch.

Jawed

PeterAce · Jan 4, 2006

With D24S8 :

Code:

                     System Information
-----------------------------------------------------------
        CPU : AMD Athlon(tm) 64 Processor 3500+
        GFX : NVIDIA GeForce 7800 GTX
         OS : Microsoft Windows XP


   Settings : 1024x768  32 bits  D24S8  No AA

                  Color Fill : 8584.063 M-Pixel/s
                      Z Fill : 17178.19 M-Pixel/s
              Color + Z Fill : 8083.263 M-Pixel/s


  Settings : 1024x768  32 bits  D24S8  2x FSAA

                  Color Fill : 8594.129 M-Pixel/s
                      Z Fill : 8596.646 M-Pixel/s
              Color + Z Fill : 8596.646 M-Pixel/s


  Settings : 1024x768  32 bits  D24S8  4x FSAA

                  Color Fill : 4572.63 M-Pixel/s
                      Z Fill : 4572.63 M-Pixel/s
              Color + Z Fill : 4308.389 M-Pixel/s

Better results!

PeterAce · Jan 4, 2006

Code:

CPU : AMD Athlon(tm) 64 Processor 3500+
        GFX : NVIDIA GeForce 7800 GTX
         OS : Microsoft Windows XP

 No AA  Settings : 1024x768    D16       D24S8

                  Color Fill : 8596.646  8584.063
                      Z Fill : 17178.19  17178.19
              Color + Z Fill : 6613.578  8083.263
     
 2x FSAA  Settings : 1024x768  D16       D24S8

                  Color Fill : 8596.646  8594.129
                      Z Fill : 8596.646  8596.646
              Color + Z Fill : 6450      8596.646

 4x FSAA  Settings : 1024x768  D16       D24S8 

                  Color Fill : 4572.63   4572.63
                      Z Fill : 4572.63   4572.63
              Color + Z Fill : 3447.718  4308.389

No AA : Colour + Z improvement = % 22
2x AA : Colour + Z improvement = % 33
4x AA : Colour + Z improvement = % 25

Jawed · Jan 4, 2006

Hmm, I wonder if that's due to:

RAM tiling of colour/z data in the backbuffer suits the 32-bit Z/stencil better than 16-bit Z
extra compressibility of stencil (missing from D16 entirely)
erm...

Hopefully we're feeding Dave with ideas for experiments.

It would be great to know if the Z-fill test is making any attempt to defeat compression or if it's trying to get maximum performance out of compression by making all values the same.

Such a wide-open topic...

Jawed

Bob · Jan 5, 2006

PeterAce said:
As expected the optimized 'Double Z when no Colour' is disabled with 2xAA or anything higher.

Seems to work just fine, looking at your numbers. 2xAA means you have 2x the number of samples. If you have double-speed Z, then you effectively run 2x AA "for free" (ignoring bandwidth) when doing Z-only rendering. Similarly, 4x AA runs at "half speed".

Pete · Jan 5, 2006

So are we seeing quadruple Z on the X1600?

Unknown Soldier · Jan 5, 2006

D16 - X800XT AGP 520/560 1024x768 @ 85 Hz with Cat 5.13
No AA
Color Fill : 5981.916 M-Pixel/s
Z Fill : 7854.254 M-Pixel/s
Color + Z Fill : 4547.464 M-Pixel/s
-----------------------------------------------------------
2xAA
Color Fill : 5903.903 M-Pixel/s
Z Fill : 6850.138 M-Pixel/s
Color + Z Fill : 4109.579 M-Pixel/s
-----------------------------------------------------------
4xAA
Color Fill : 5496.215 M-Pixel/s
Z Fill : 3925.869 M-Pixel/s
Color + Z Fill : 3379.77 M-Pixel/s

**********************

D24S8 - X800XT AGP 520/560 1024x768 @ 85 Hz with Cat 5.13
No AA
Color Fill : 5946.685 M-Pixel/s
Z Fill : 8088.295 M-Pixel/s
Color + Z Fill : 4570.114 M-Pixel/s
-----------------------------------------------------------
2xAA
Color Fill : 5896.353 M-Pixel/s
Z Fill : 7439.018 M-Pixel/s
Color + Z Fill : 4124.678 M-Pixel/s
-----------------------------------------------------------
4xAA
Color Fill : 5476.083 M-Pixel/s
Z Fill : 3928.385 M-Pixel/s
Color + Z Fill : 3193.543 M-Pixel/s

FrameBuffer · Jan 5, 2006

don't know if this helps anyone:

first set from ATI X800 XT AIW AGP (500/500)

System Information
-----------------------------------------------------------
CPU : Dual Core AMD Opteron(tm) Processor 165
GFX : ALL-IN-WONDER X800 Series
OS : Microsoft Windows XP

Benchmark Result
-----------------------------------------------------------

1024 NO FSAA 16bit D16bit
Color Fill : 2851.288 M-Pixel/s
Z Fill : 3553.414 M-Pixel/s
Color + Z Fill : 2239.758 M-Pixel/s

1024 NO FSAA 32bit D16bit
Color Fill : 4972.767 M-Pixel/s
Z Fill : 3757.258 M-Pixel/s
Color + Z Fill : 3040.032 M-Pixel/s

1024 2x FSAA 16bit D16bit
Color Fill : 2843.738 M-Pixel/s
Z Fill : 3550.898 M-Pixel/s
Color + Z Fill : 2249.825 M-Pixel/s

1024 2x FSAA 32bit D16bit
Color Fill : 4970.25 M-Pixel/s
Z Fill : 3744.675 M-Pixel/s
Color + Z Fill : 3050.098 M-Pixel/s

1024 4x FSAA 16bit D16bit
Color Fill : 2853.804 M-Pixel/s
Z Fill : 3553.414 M-Pixel/s
Color + Z Fill : 2267.441 M-Pixel/s

1024 4x FSAA 32bit D16bit
Color Fill : 4962.7 M-Pixel/s
Z Fill : 3749.708 M-Pixel/s
Color + Z Fill : 3045.065 M-Pixel/s

---------------------------------------

1024 NO FSAA 16bit D24S28
Color Fill : 2831.155 M-Pixel/s
Z Fill : 3550.898 M-Pixel/s
Color + Z Fill : 2277.507 M-Pixel/s

1024 NO FSAA 32bit D24S28
Color Fill : 4970.25 M-Pixel/s
Z Fill : 3739.642 M-Pixel/s
Color + Z Fill : 2878.97 M-Pixel/s

1024 2x FSAA 16bit D24S28
Color Fill : 2846.255 M-Pixel/s
Z Fill : 3555.931 M-Pixel/s
Color + Z Fill : 2317.772 M-Pixel/s

1024 2x FSAA 32bit D24S28
Color Fill : 4857.004 M-Pixel/s
Z Fill : 3754.741 M-Pixel/s
Color + Z Fill : 2929.302 M-Pixel/s

1024 4x FSAA 16bit D24S28
Color Fill : 2851.288 M-Pixel/s
Z Fill : 3558.447 M-Pixel/s
Color + Z Fill : 2290.09 M-Pixel/s

1024 4x FSAA 32bit D24S28
Color Fill : 4972.767 M-Pixel/s
Z Fill : 3747.191 M-Pixel/s
Color + Z Fill : 2889.037 M-Pixel/s

FrameBuffer · Jan 5, 2006

hmm .. not sure if it's a driver issue, setting or application however in ALL tests with FSAA enabled no matter what color setting or format the ATI x1600 exhibits an anomoly during the 2nd of the 3 tests, the screen shows as though the Hsynch is out of whack,.. like when trying to adjust an old color TV.. anyhow:

System Information
-----------------------------------------------------------
CPU : Dual Core AMD Opteron(tm) Processor 165
GFX : RADEON X1600 Series Secondary
OS : Microsoft Windows XP

Benchmark Result
-----------------------------------------------------------
1024 NO FSAA 16bit D16
Color Fill : 2212.076 M-Pixel/s
Z Fill : 4655.677 M-Pixel/s
Color + Z Fill : 2212.076 M-Pixel/s
1024 NO FSAA 32bit D16
Color Fill : 2212.076 M-Pixel/s
Z Fill : 4653.161 M-Pixel/s
Color + Z Fill : 2212.076 M-Pixel/s

1024 2x FSAA 16bit D16
Color Fill : 1930.219 M-Pixel/s
Z Fill : 3656.594 M-Pixel/s
Color + Z Fill : 1804.39 M-Pixel/s
1024 2x FSAA 32bit D16
Color Fill : 2040.948 M-Pixel/s
Z Fill : 4084.413 M-Pixel/s
Color + Z Fill : 1889.953 M-Pixel/s

1024 4x FSAA 16bit D16
Color Fill : 1882.404 M-Pixel/s
Z Fill : 2061.081 M-Pixel/s
Color + Z Fill : 1560.281 M-Pixel/s
1024 4x FSAA 32bit D16
Color Fill : 1988.1 M-Pixel/s
Z Fill : 2191.943 M-Pixel/s
Color + Z Fill : 1623.196 M-Pixel/s

1024 NO FSAA 16bit D24S8
Color Fill : 2212.076 M-Pixel/s
Z Fill : 4660.71 M-Pixel/s
Color + Z Fill : 2212.076 M-Pixel/s
1024 NO FSAA 32bit D24S8
Color Fill : 2212.076 M-Pixel/s
Z Fill : 4655.677 M-Pixel/s
Color + Z Fill : 2212.076 M-Pixel/s

1024 2x FSAA 16bit D24S8
Color Fill : 1930.219 M-Pixel/s
Z Fill : 3523.215 M-Pixel/s
Color + Z Fill : 1713.792 M-Pixel/s
1024 2x FSAA 32bit D24S8
Color Fill : 2040.948 M-Pixel/s
Z Fill : 3913.286 M-Pixel/s
Color + Z Fill : 1789.29 M-Pixel/s

1024 4x FSAA 16bit D24S8
Color Fill : 1882.404 M-Pixel/s
Z Fill : 2061.081 M-Pixel/s
Color + Z Fill : 1545.182 M-Pixel/s
1024 4x FSAA 32bit D24S8
Color Fill : 1988.1 M-Pixel/s
Z Fill : 2194.46 M-Pixel/s
Color + Z Fill : 1603.063 M-Pixel/s

Also unable to test any of the PS2.0 tests beyond the 2.0 PP Simple ..

Jawed · Jan 5, 2006

There's a wide disparity in X800 performance comparing 16- and 32-bit colour but relatively little in X1600XT.

That would seem to indicate that X1600XT's more granular memory access are making it more tolerant. Or it may indicate that X1600XT arranges colour and Z/stencil data in a different tiling in memory.

Or both?

Someone needs to write a backbuffer snooping application, or tweak these fillrate testing apps to draw out a memory map for the way that the backbuffer is organised.

Jawed

X1600 MSAA Fillrates

Unknown Soldier

PeterAce

Jawed

PeterAce

PeterAce

Jawed

Bob

Pete

Moderate Nuisance

Unknown Soldier

FrameBuffer

FrameBuffer

Jawed

Similar threads