Hi Yoda Master,
Thanks for your response.
Yes I am using Dark GDK .... but i program in "C" mode rather than "OOP" style as I believe that structured programing produces faster code.
I am starting to think that I jumped the gun about what the actual problem is.
Basically I am creating a world in 3 layers. My world happens to cover an area of 1000KM x 1000KM.
I am using 2 active cameras (camera 0 covers distance from 0.3 to 300 meters, and camera 1 covers from 300 to 300,000 meters) The images are merged to create the final view screen. This was necessary to resolve Z fighting problems.
It has an ocean bed covering that area and made up of a matrix of 128x128.
It has a transparent ocean of the same area, and also made up of a matrix of 128x128.
It has a land mass going from just below the ocean bed level, cutting through the ocean level and poking into the air. The land mass is also made of a matrix of 128x128. the land mass is skinned with a 4096x4096 pixel texture with a nice picture added into it to test for resolution capability of the land mass. Here is a screen shot.
By removing one of the meshes on my problem machine, the FPS jumps back to normal .... and as it was the machine with the smallest cache memory, I assumed it was a CPU cache problem.
After double checking, it is also the machine with an ATI X300se as the video card. All the other machines are running ATI HD2400 or better video cards. After having other people test the program with different generation CPU, even down to old slow Socket A athlons, the common thread for this FPS collapse is looking to be Video card related rather than CPU cache memory related.
When others are using things like ATI 9800 Pro, and that earlier generation of video cards, then the collapse in FPS happens.
It is looking more like a problem when meshes intersect (land mass intersecting through 2 other meshes) that may be the real problem with these earlier video cards.
I have confirmed, that there is no noticeable difference in performance between using *.X based models (mesh based) for the terrain etc, compared to using Matrix constructions.
I still want to do some more testing when I get the time.
The reason for this elaborate explanation post is that should people at a later date get back to this post, I hope they will not take my first post in this thread as being accurate. I think I did a bit too much assuming, and it looks like I was wrong.
Regards sydbod