Reduce graphic resource allocation calls

[kennyvv]

The Issue

We create too many vertexbuffers, for example every single entity (yes, every entity. not just every model) gets its own vertexbuffer.
This is obviously very wasteful, any operation involving the graphicsdevice needs to be ran on the UI thread. This means that everytime we create or update a vertexbuffer, indexbuffer or texture we need queue that task up on the UI thread. This slows down the render/update loop a considerable amount, creating lag spikes or even causing the game to freeze completely.

The Solution

Entity & Item models

For models we could start using model instancing, this means we only create 1 vertexbuffer for every model. We can then use that same vertexbuffer for every instance of that model we need to draw. Exceptions to this would be when a player has unique geometry data. In this case we would still need to create a new buffer just for that player.

Chunks

Instead of creating a new vertexbuffer for each individual chunk we could also create 1 world vertexbuffer (Let's say we allocate 1gb of gpu memory) and allocate sections of that to each chunk. This does come with the overhead of having to keep track of allocated sections however. I'm also not sure what impact this will actually have on performance. This also does not solve the fact that we still need to send buffer updates to the gpu, it will however solve the issue of having to create a new vertexbuffer everytime the size of a chunks vertexbuffer increases (by placing blocks for example).

Another thing i have been thinking about is how we can keep track of updated vertices in a chunk, so when a user places a block for example we only send the new vertices of that block to the gpu instead of having to recalculate and resend the whole vertex array of that chunkcolumn to the gpu.