RPTD

I would recommend always performing the creation of mip-maps and the compression of your texture data at data-build time, instead of at run-time.
There are many libraries to help with this task, such as:
http://code.google.c...-texture-tools/
http://code.google.com/p/libsquish/
http://developer.amd...es/default.aspx

Figured that much out myself too. Looks like I have to do the compression software side. I had this plan anyways so that's not much of a bummer. Downloaded that libsquish already but could not yet put it to use. Coming soon.

@vNeeki:
Create all mip map levels in software either using your own software or some other software. Since it's down-sampling by 2 (hence 4 pixels into 1) you can use a simple box filter for that. Once done you can upload each level using glTexImage2D specifying 0, 1, ... , N as the first parameter to write the appropriate mip map level.

That's like magic... whenever I post here a question after weeks of looking for an answer I stumble across the solution soon after.

Figured out what causes the problem. I disabled mip mapping while using compression (hence GL_COMPRESSED_RGB_ARB respectivly GL_COMPRESSED_RGB_ARBA) and then the initial delay vanished. So the conclusion is that if a compression format is used with mip mapping ATI doesn't compress the data while using glTexImage2D but delays the compression of the data until the texture is rendered for the first time. With either compression disabled or mip mapping disabled the delay vanished. So you can't do both (compression and mip mapping) on ATI without running into the delay problem.

if would perhaps refine rule 2 a bit. it is ok and often very helpfull if you know the basic gameplay you aim at before you go into the story. it does not help much to think of characters and their abilities if the game later on doesn't honor this. i would think about both in the same time instead of one by one.

I'm trying to optimize my render code. During this I noticed that the copying values to the VBO takes quite some time. The VBO is a dynamic one as the mesh bends around ( creature ). Now I question myself if I am quicker using a memory-array instead of making a VBO. I also would like to keep in mind the memory consumption. With higher resolution meshes the VBO data can quickly explode eating precious texture memory. Is it worth dropping a VBO in favor of cpu memory especially if the data changes every frame?