http://service.madonion.com/compare?2k1=4202539
I liked to see the detailed test result. Very impressive.
I liked to see the detailed test result. Very impressive.
R300 3DMARK2001 PROJECT COMPARE
- Thread starter pascal
- Start date
Randell said:
3DMark Score 17421 3D marks
Game 1 Car Chase - Low Detail 228.8 FPS
Game 1 Car Chase - High Detail 85.9 FPS
Game 2 Dragothic - Low Detail 325.1 FPS
Game 2 Dragothic - High Detail 171.7 FPS
Game 3 Lobby - Low Detail 227.9 FPS
Game 3 Lobby - High Detail 101.9 FPS
Game 4 Nature 120.6 FPS
Fill Rate (Single-Texturing) 2013.3 MTexels/s
Fill Rate (Multi-Texturing) 2848.8 MTexels/s
High Polygon Count (1 light) 79.1 MTriangels/s
High Polygon Count (8 lights) 16.9 MTriangels/s
Environment Bump Mapping 212.1 FPS
DOT3 Bump Mapping 258.1 FPS
Vertex Shader 219.5 FPS
Pixel Shader 235.0 FPS
Advanced Pixel Shader 216.1 FPS
Point Sprite 42.0 MSprites/s
TheMightyPuck
Newcomer
i think 3dmark is pretty much cpu these days
pascal said:
You are right if this is any indication.
http://service.madonion.com/compare?2k1=3990219
TheMightyPuck
Newcomer
holy howling hell that baby is oc'd off the charts
Why are the point sprites performing much worse than triangles with 1 light?
Someone else asked what point sprites are useful for? Point sprites are billboard objects. They are quads that always face forward towards the viewer and they have position and size only.
So what are they useful for? Two Words: Particle Systems. Explosions, beams, fog, smoke, fire, fountains, trails, etc
The reason they are useful is that they are cheaper than triangles to use. All you need is (X,Y,Z) and a size, you don't need to do any perspective correction, their normal is constant, etc. If the R300 can achieve near 80MT/s with 1 light, it should be able to do over double that amount in particles/point sprites.
Something is either wrong with the driver, or wrong with 3DMark. It could be that 3dMark's point sprite test is CPU limited.
Someone else asked what point sprites are useful for? Point sprites are billboard objects. They are quads that always face forward towards the viewer and they have position and size only.
So what are they useful for? Two Words: Particle Systems. Explosions, beams, fog, smoke, fire, fountains, trails, etc
The reason they are useful is that they are cheaper than triangles to use. All you need is (X,Y,Z) and a size, you don't need to do any perspective correction, their normal is constant, etc. If the R300 can achieve near 80MT/s with 1 light, it should be able to do over double that amount in particles/point sprites.
Something is either wrong with the driver, or wrong with 3DMark. It could be that 3dMark's point sprite test is CPU limited.
Interesting... http://service.madonion.com/compare?2k1=3566685
A heavily o/ced GF4 gets 64MTriangels/s and 37.5MSprites/s, or ~0.6 Sprite/Triangel. Although it is slighly higher then 9700s 0.53 S/T, it is still nowhere near 1/1.
On the other hand... http://service.madonion.com/compare?2k1=3299635
This o/ced GF3 gets 0.94 S/T
http://service.madonion.com/compare?2k1=4102980
Here, Radeon 8500 gets 0.7 S/T
It seems that the faster the card, the lower the Sprite/Triangle ration:
GF3 > R8500 > GF4 > R9700.
A heavily o/ced GF4 gets 64MTriangels/s and 37.5MSprites/s, or ~0.6 Sprite/Triangel. Although it is slighly higher then 9700s 0.53 S/T, it is still nowhere near 1/1.
On the other hand... http://service.madonion.com/compare?2k1=3299635
This o/ced GF3 gets 0.94 S/T
http://service.madonion.com/compare?2k1=4102980
Here, Radeon 8500 gets 0.7 S/T
It seems that the faster the card, the lower the Sprite/Triangle ration:
GF3 > R8500 > GF4 > R9700.
OpenGL guy
Veteran
Whoa! Thanks for the idea! I'm making the change to our driver right now!DemoCoder said:
But seriously, you aren't taking the datasets into consideration. For example, how large are the triangles in the hi-poly 1 light test vs. the size of the point sprites in the point sprite test?
As soon as you analyze the data, let us know.
What investigation have you done that leads you to believe there's a possible driver issue here? Maybe Geeforcer is on to something with his data.
