Q. Triangle Batching with older cards (GF2)
Blitz3D Forums/Blitz3D Programming/Q. Triangle Batching with older cards (GF2)
| ||
| Mustang mentioned that 2000 polys per surface is what current Video cards are optimised to handle. Could he, or anyone else, give an estimate of what value would apply to an older card like a Geforce 2 (MX)? Purely out of self interest as that is the card I am using and would like to get an idea for what it should be able to handle as well as how to organise my media for games I make to run on my own PC. |
| ||
| I think it's directx that batches the triangles like that. Basically keeps your triangles at 2000 at the most for each surface. I did alot of tests on that, on different computers and they seemed to yield the same result. Small FPS increases, everytime i added more triangles to a surface. When i hit > 2000, the FPS was slightly greater than the FPS that came before hand. |
| ||
| ..And that 2K is MINIMUM optimial size - the bigger the better of course. It's always beneficial to build very large polygroups that use the same surface because modern HW likes to render large amount of similar things... doing lot of small batches makes them stall and not hit their best performance. All this within reason of course, beacuse breaking meshes to smaller pieces will speed up rendering throug object culling. It's quite tricky to do "optimized" 3D beacuse the optimum case is a result of so many variables. And the batch size thing comes from the HW manufacturers who have been optimizing their HW for years for large poly batches ("our blaablaa can render quantizillionpolygons per cycle"). Also modern games DO have ridiculous amount of polygons to render, one character is easily between 5000 - 10000 polgons, and this has additional with specular and bump maps etc which of course multiply the rendering work (multiple render passes). |
| ||
| Thanks |
| ||
| Are you saying that: - 5 surfaces with 10 triangles each IS SLOWER THAN: - 5 surfaces with 2000 triangles each ???? Why? |
| ||
| No, what he's saying is, the speed difference between the two isn't much at all. So, you'd be as well filling up the surface with triangles to optimise speed :o) |
| ||
| Yes, what Ross said. It's like if you'd have a nice Porsche for example - we all know that they are fast but getting into the car, starting it and fastening the seatbelt takes time. After that it really doesn't matter if you want to acclerate to 50mph or 150mph, difference is minimal compared to the time taken before you can push the pedal to the metal. Same analogy goes with 3D cards, because setup phase(s) takes so much time compared to the actual pushing of polygon (welll vertex) data through the pipes and rendering the screen... and changing the surface? Think of it as stopping the Porsche, getting out of it, getting back in, fastening the seatbelt... |
   |