VBO not getting data

Hello,

I recently added VBO support to a project and I am having some issues with different NVIDIA cards. The issue that I am having is it seems on certain cards, some objects are not even making it into the buffer.

I’m loading the data from a custom OBJ into a few lists depending on certain properties of the objects. This code all works and everything displays fine on every compatible machine. One of my lists contains a large amount of static geometry that I am attempting to put into a VBO. My process is as follows.

Setting up the buffer

void vbo_manager::init(range_t range)
{
    unsigned int size = 0;
    unsigned int offset = 0;

    glGenBuffers(1,&vbo_id);
    glBindBuffer(GL_ARRAY_BUFFER,vbo_id);
    //determine total size of buffer
    for( prim_t::iterator pit = range.first; pit != range.second; ++pit)
    {
        size += pit->getVertexCount();
    }

    //set up buffer area
    glBufferData(GL_ARRAY_BUFFER,sizeof(vertex)*size,NULL,GL_STATIC_DRAW);

    size = 0;
    //copy data to buffer
    for( prim_t::iterator pit = range.first; pit != range.second; ++pit)
    {
        size = pit->getVertexCount*sizeof(vertex);
        glBufferSubData(GL_ARRAY_BUFFER,offset,size,(void*)pit->getAddressOfData() );
        offset += size;
    }

    glBindBuffer(GL_ARRAY_BUFFER,0);
}

On certain machines, portions of the VBO don’t get mapped. This seems to occur whenever the size parameter of the glBufferSubData() increases beyond a certain amount or the offset and size interact in some strange way. If I try and retrieve the data with glGetBufferSubData() at a location that is having trouble I receive all 0’s.

The trouble has shown up on three different machines running in Windows with Nvidia cards: a notebook with 9800M GTS, a desktop with 275gtx, and a desktop with 460gtx, all with up to date drivers (275.xx). On another desktop with a 280gtx and on any of the machines in Linux, the problem does not exist.

I can get the triangles to reappear if I split larger objects into multiple glBufferSubData calls with smaller size parameters, or switch to using glMapbuffer. Is there some hard limit on glBufferSubData that I can not find in documentation or is there something more sinister happening?