Vertex Array Object vs. Vertex Buffer Object

What is the relationship between a VAO and a VBO?

VBOs manage vertex buffers (possibly) on the GPU.

VAOs manage client side state, i.e. VBO bindings, IBO bindings and vertex layout.

For instance you can bind one VAO, then bind a VBO, an IBO and configure the vertex layout, bind another VAO, another VBO and IBO
and configure a different vertex layout. Whenever you bind the first VAO, the VBO and IBO asociated with it are bound and the asociated
vertex format is used. Binding the second VAO will then switch to the other pair of VBO/IBO and respective vertex layout.

A “buffer” or “buffer object” is a block of memory which may reside in video memory, meaning that you can’t simply grab a pointer to it, write into it, and expect the changes to automatically affect subsequent operations. You need to either explicitly copy data into it (e.g. glBufferSubData), or map it to the CPU’s address space (e.g. glMapBuffer) then unmap it before instructing the GPU to access it (it is an error to execute a GL operation which reads or writes a buffer while it is mapped).

A vertex buffer object (VBO) is a buffer which is used to hold a vertex attribute array (aka vertex array). If a buffer is bound to the GL_ARRAY_BUFFER binding point when glVertexAttribPointer (or glVertexPointer, glTexCoordPointer etc in the compatibility profile) is called, the “pointer” argument is interpreted as an offset into the buffer rather than as a pointer to client memory. Subsequent draw calls (glDrawArrays, glDrawElements, etc) will read data attribute data from the buffer rather than from client memory. The buffer used is the one bound at the time of the glVertexAttribPointer (etc) call, not the draw call.

A vertex array object (VAO) is a named object used to group together certain state relating to vertex arrays. If a VAO is bound (via glBindVertexArray), functions which affect how vertex attributes are sourced by the vertex shader modify the state in the bound VAO rather than global state. Specifically, the state stored in a VAO is, for each vertex array:

[ul]
[li] The size, type, stride, and normalized flag as set by glVertexAttribPointer.
[/li][li] The “pointer” (i.e. offset) as set by glVertexAttribPointer.
[/li][li] The buffer holding the data, i.e. that bound to GL_ARRAY_BUFFER at the time of the last glVertexAttribPointer call.
[/li][li] Whether the array contains integers (glVertexAttribIPointer) or fixed/floating point values (glVertexAttribPointer).
[/li][li] the divisor, as set by glVertexAttribDivisor.
[/li][li] whether the array is enabled (glEnableVertexAttribArray, glDisableVertexAttribArray).
[/li][/ul]
In addition to the above per-array data, the buffer currently bound to GL_ELEMENT_ARRAY_BUFFER is stored in the VAO.

Note that the primitive restart index and primitive restart enabled state isn’t stored in a VAO although (conceptually) it probably should be.

By grouping all of this state into a VAO, you can change all of it with a single glBindVertexArray call, rather than having to make 4 calls (glBindBuffer, glVertexAttribPointer, glVertexAttribDivisor, gl{Enable,Disable}VertexAttribArray) per vertex array plus one more for the element array.

All the tutorials I have read start with one primitive (i.e. a triangle) and render it. I have this idea to create a simple verions of the asteroids game. I will have one spaceship (a triangle) and ten asteroids (hexagons). Would I reuse my vao and vbo variables for each shape or have a separate one for each shape?

For something that simple, it doesn’t really matter. You could draw everything with a single VBO, no VAO, and a single draw call.

For a more complex program, you’d put data with different access patterns in different buffers (you need to supply usage hints for the buffer as a whole). Buffers which will be used together would be grouped using VAOs to minimise the total number of calls required to set up for each draw call.