noko
Distinguished
<b>Quake III</b>, 1024x768x32 max everything, non compressed textures, version 127h timedemo1, demo demo127
Radeon 64 VIVO Retail, IWILL KK266, T-Bird @ 866mhz, 512mb Ram, Driver 3132 Special Purpose, W2K SP 1.
. <b>Vertex lighting</b>
. . . Bi - 90.7FPS
. . . Tri - 81 FPS (Better visual quality)
. . . Max Anisotropic 74.9FPS (Great increase in visual quality)
*<i>By design the Kyro2 would have to drop at least 1/2 the FPS from Tri to max anisotropic filtering.</i>
. <b>Light table</b>(QuakeIII looks better here)
. . . Bi - 71.5FPS
. . . Tri - 64.6FPS
. . . Max Anisotropic 59.7FPS (Awesome visual quality, this is where I play.)
<i>Note: Loose about 5FPS from going from Tri to Anisotropic filtering. Add about 10FPS when using compressed textures. I don't use them due to the better quality with no apparent loss in game play for me. Plus I normally use the next resolution 1152x864x32 anyways.</i>
<b>Serious Sam</b>, version 1.0, 1024x768x32, Execute Addon - GFX: Extreme Quality (max everything),Multitexturing was at Triple (meaning using the 3 texture units of the Radeon) with the following modifications: Precache Shadowmaps - Yes(On), moved Cache size from large to huge.
<i>Note: With W2k and the above configuration it takes close to 400mb of physical Ram.</i>
. <b>Serious Sam Benchmarks</b> - dem_bProfile=1, Dunes
. . . Bi - 66.8FPS (anisotropic turned off in Drivers)
. . . Tri - 62.2FPS (anisotropic turned off in Drivers)
. . . Standard Anisotropic 65FPS (anisotropic Turned On in drivers as well as in game)
. . . Max Anisotropic 63.3FPS (anisotropic Turned On in drivers as well as in game)
Not sure of the results, did see improvement from Bi to Tri and to Anisotropic but the improvement is no way as near the dramatic changes that I saw in QuakeIII. Downloaded 1.OOc patch and installed after the above tests. Will test later and post if significantly different.
The comment made that the Radeon can't use the third texture unit when doing anisotropic filtering is incorrect and shows the ignorance of the reviewer. The texture unit has nothing to do with anisotropic filtering. The filtering is done before it is sent to the texture unit.
PowerVR created the VilliageMark benchmark to make the Geforce look bad (my opinion) and the Kyro to look good. It renders front to back which is not optimum for a traditional renderer but great for the Radeon HierarchialZ. When T&L is turned off on a T&L card the performance goes up!! over 15% alone. Yet if you have a T&L card villiageMark inserts its T&L optimization that hurts the performance while on the Kyro it removes its hindering method. Its radical results from Bi to Tri with the Radeon is not reflected in real world applications so as for a benchmark it is virtually worthless. Now the test done at Ace's Hardware apparently had HierarchialZ turned off in which some driver sets does have hierarchialZ turned off which is easy to turn on. Once again my benchmarks in VilliageMark 1.17 with 3102 drivers in W2K.
<b>VilliageMark, 1024x768x16</b>
. . . T&L on, Trilinear On, hierarchialZ off - 40FPS
. . . T&L on, Trilinear On, hierarchialZ on - 51FPS (<i>Notice the jump, it renders front to back</i>)
. . . T&L off, Trilinear On, hierarchiaZ on - 58FPS (<i>hmmmmmmmmmmmm those dirty cheap....</i>)
. . . T&L off, Bilinear On, hierarchiaZ on - 86FPS (<i><b>Smoking!!!</i></b>)
The Kyro2 anisotropic filtering imposes automatically 1/2 the framerate at a minimum. In short the Kyro2 anisotropic filtering <b>SUCKS</b> on degrading performance when actually used. Its quality is on par with the Radeon doing 16tap.
I hope this clears up Anisotropic filtering on the Radeon somewhat. PowerVR how about doing some benchmarks and posting them yourself using your Kyro. QuakeIII, Serious Sam, VilliageMark using Bi, Tri and Anisotropic. Use compress textures on the Kyro because I know it is 32mb. QuakeIII seems to be the best benchmark for doing filtering method testing.
<P ID="edit"><FONT SIZE=-1><EM>Edited by noko on 05/01/01 11:43 AM.</EM></FONT></P>
Radeon 64 VIVO Retail, IWILL KK266, T-Bird @ 866mhz, 512mb Ram, Driver 3132 Special Purpose, W2K SP 1.
. <b>Vertex lighting</b>
. . . Bi - 90.7FPS
. . . Tri - 81 FPS (Better visual quality)
. . . Max Anisotropic 74.9FPS (Great increase in visual quality)
*<i>By design the Kyro2 would have to drop at least 1/2 the FPS from Tri to max anisotropic filtering.</i>
. <b>Light table</b>(QuakeIII looks better here)
. . . Bi - 71.5FPS
. . . Tri - 64.6FPS
. . . Max Anisotropic 59.7FPS (Awesome visual quality, this is where I play.)
<i>Note: Loose about 5FPS from going from Tri to Anisotropic filtering. Add about 10FPS when using compressed textures. I don't use them due to the better quality with no apparent loss in game play for me. Plus I normally use the next resolution 1152x864x32 anyways.</i>
<b>Serious Sam</b>, version 1.0, 1024x768x32, Execute Addon - GFX: Extreme Quality (max everything),Multitexturing was at Triple (meaning using the 3 texture units of the Radeon) with the following modifications: Precache Shadowmaps - Yes(On), moved Cache size from large to huge.
<i>Note: With W2k and the above configuration it takes close to 400mb of physical Ram.</i>
. <b>Serious Sam Benchmarks</b> - dem_bProfile=1, Dunes
. . . Bi - 66.8FPS (anisotropic turned off in Drivers)
. . . Tri - 62.2FPS (anisotropic turned off in Drivers)
. . . Standard Anisotropic 65FPS (anisotropic Turned On in drivers as well as in game)
. . . Max Anisotropic 63.3FPS (anisotropic Turned On in drivers as well as in game)
Not sure of the results, did see improvement from Bi to Tri and to Anisotropic but the improvement is no way as near the dramatic changes that I saw in QuakeIII. Downloaded 1.OOc patch and installed after the above tests. Will test later and post if significantly different.
The comment made that the Radeon can't use the third texture unit when doing anisotropic filtering is incorrect and shows the ignorance of the reviewer. The texture unit has nothing to do with anisotropic filtering. The filtering is done before it is sent to the texture unit.
PowerVR created the VilliageMark benchmark to make the Geforce look bad (my opinion) and the Kyro to look good. It renders front to back which is not optimum for a traditional renderer but great for the Radeon HierarchialZ. When T&L is turned off on a T&L card the performance goes up!! over 15% alone. Yet if you have a T&L card villiageMark inserts its T&L optimization that hurts the performance while on the Kyro it removes its hindering method. Its radical results from Bi to Tri with the Radeon is not reflected in real world applications so as for a benchmark it is virtually worthless. Now the test done at Ace's Hardware apparently had HierarchialZ turned off in which some driver sets does have hierarchialZ turned off which is easy to turn on. Once again my benchmarks in VilliageMark 1.17 with 3102 drivers in W2K.
<b>VilliageMark, 1024x768x16</b>
. . . T&L on, Trilinear On, hierarchialZ off - 40FPS
. . . T&L on, Trilinear On, hierarchialZ on - 51FPS (<i>Notice the jump, it renders front to back</i>)
. . . T&L off, Trilinear On, hierarchiaZ on - 58FPS (<i>hmmmmmmmmmmmm those dirty cheap....</i>)
. . . T&L off, Bilinear On, hierarchiaZ on - 86FPS (<i><b>Smoking!!!</i></b>)
The Kyro2 anisotropic filtering imposes automatically 1/2 the framerate at a minimum. In short the Kyro2 anisotropic filtering <b>SUCKS</b> on degrading performance when actually used. Its quality is on par with the Radeon doing 16tap.
I hope this clears up Anisotropic filtering on the Radeon somewhat. PowerVR how about doing some benchmarks and posting them yourself using your Kyro. QuakeIII, Serious Sam, VilliageMark using Bi, Tri and Anisotropic. Use compress textures on the Kyro because I know it is 32mb. QuakeIII seems to be the best benchmark for doing filtering method testing.
<P ID="edit"><FONT SIZE=-1><EM>Edited by noko on 05/01/01 11:43 AM.</EM></FONT></P>