Tell me about polygoncount
- Thread starter phed
- Start date
DeathKnight
Regular
I don't see how it's misleading. It's still a count of the polygons that are being drawn.zurich said:
The EA results are reasonably accurate..
The model is stripped, so extra vertices are set ( The vertex count for the bunny is closer to the PS2 specification sheet released publically - it's not flat shaded )
On the Xbox the post-transform cache artificially increases the vertex count, as reused vertices often bypass transformation completely - and the method of 'welding' small strips together involves creating degenerate triangles ( which boost the vertex count compared to the polygon count )
As a benchmark of throughput the EA paper is quite comprehensive, and they go into a bit of detail about the mechanisms.
The model is stripped, so extra vertices are set ( The vertex count for the bunny is closer to the PS2 specification sheet released publically - it's not flat shaded )
On the Xbox the post-transform cache artificially increases the vertex count, as reused vertices often bypass transformation completely - and the method of 'welding' small strips together involves creating degenerate triangles ( which boost the vertex count compared to the polygon count )
As a benchmark of throughput the EA paper is quite comprehensive, and they go into a bit of detail about the mechanisms.
Highest polygon counts I know....
Dreamcast- 5-6 million in game, 10 million in tech demo.
Gamecube- 15 million in game, 20-25 million with I think 8 texture passes during a benchmark
PS2- around 12 million in game, 15 to 20 million in a benchmark
Xbox- 20-25 million with 4 texture passes during a benchmark(could be using the wrong phrase here, it may not be texture passes)
Dreamcast- 5-6 million in game, 10 million in tech demo.
Gamecube- 15 million in game, 20-25 million with I think 8 texture passes during a benchmark
PS2- around 12 million in game, 15 to 20 million in a benchmark
Xbox- 20-25 million with 4 texture passes during a benchmark(could be using the wrong phrase here, it may not be texture passes)
PS2 benchmark
http://playstation2-linux.com/projects/sps2
In the vfast sampe directory ( Anyone can download it ) is a benchmark program for PS2 Linux that draws 34M polys/second.
That doesnt use the core cpu or vu0...
http://playstation2-linux.com/projects/sps2
In the vfast sampe directory ( Anyone can download it ) is a benchmark program for PS2 Linux that draws 34M polys/second.
That doesnt use the core cpu or vu0...
One nice thing about theoretical max. numbers is that they are mostly written down by the manufacturers themselves, so we don't have to figure as much 
Anyway, to be exact, it can do 16texture accesses without resending geometry (with 8 different textures). But there are other things that eat geometry time - one of them being UV setup in this case.
Anyway, to be exact, it can do 16texture accesses without resending geometry (with 8 different textures). But there are other things that eat geometry time - one of them being UV setup in this case.
David_South#1
Banned
I'm just adding this poly-fact.
In game play with all the other stuff going on in real time:
The graphics engine of GT4 reliably displays up to 6 Million polygons a second.
Fox5 has the numbers about right.
Depending on the game and the amount of other process the spec will change.
Even without a processing limit. Display models need only so many polygons.
So with extra room to process the number of polygons in a game does not reflect raw ability.
In game play with all the other stuff going on in real time:
The graphics engine of GT4 reliably displays up to 6 Million polygons a second.
Fox5 has the numbers about right.
Depending on the game and the amount of other process the spec will change.
Even without a processing limit. Display models need only so many polygons.
So with extra room to process the number of polygons in a game does not reflect raw ability.
Fafalada said:One nice thing about theoretical max. numbers is that they are mostly written down by the manufacturers themselves, so we don't have to figure as much
Anyway, to be exact, it can do 16texture accesses without resending geometry (with 8 different textures). But there are other things that eat geometry time - one of them being UV setup in this case.
In the case of the PS2, how much(numerically) the multi pass affect the polygonal throughput?
randycat99
Veteran
I guess there is some consistency if you consider that the 250k scenes may be one of those few instances where it drops to 30 fps, which would give a 7.5 MPoly/sec figure.
marconelly!
Veteran
The maximum in-game on PS2 that I know of is 18MPPS as quoted by the Melbourne House programmers, for their F1 game.
Ok, I made a bummer.
The NGC number I quoted was for 200mhz Flipper part(last week I was still sorely lacking sleep, sorry) - so the real number is in fact around 5.5M.
In real world complex multipass shaders will be far more demanding on geometry then just simplest UV setup though, so that can become a limit, as well as fillrate in some situations.
Somewhat similar rule applies to GC where afaik you'd be bound by fillrate and input streams more often then theoretical output of the geometry processor.
The NGC number I quoted was for 200mhz Flipper part(last week I was still sorely lacking sleep, sorry) - so the real number is in fact around 5.5M.
And GT3 runs 20M too, I mean it must be true it came straight from Kaz's mouth.Tagrineth said:
You'd be limited by triangle setup in theory.AbbA said:
In real world complex multipass shaders will be far more demanding on geometry then just simplest UV setup though, so that can become a limit, as well as fillrate in some situations.
Somewhat similar rule applies to GC where afaik you'd be bound by fillrate and input streams more often then theoretical output of the geometry processor.
