I''ve read some article on web considering size of vertex arrays, it was for directx though. Anyway, question is is it better to merge all arrays into one huge and then draw it or still keep array split? What size is prefered?

Wasn't it about GDC2003 and CPU bounding problems? Oh well, it does not matter.

The point is the more vertices you render at one time, the less time they'll take (proportionally).
Now, merging everything in a single array **may** give you a nice speedup but... let's face it. Doing it correctly is going to be hell on earth.
Take for example NV_VAR and the new ARB_buffer_object.
VAR tells basically that you have to allocate a large array and 'break it down' to sub-arrays. Changing from one array to the other is **slow**.
ARB_buffer_object by the other hand, was designed to change between different arrays with ease.

Now, what we have to learn from this?
Excessive optimization does not pay.
Faster development pays.

EDIT: sum up - Keep arrays separate. The best performance is implementation dependant and application-dependant.

[edited by - Krohm on May 26, 2003 3:10:07 PM]