The
version of Geometry clipmaps that appeared at Siggraph this year had most of
its components implemented on the CPU. |
|
The
current system has most of its components on the GPU. So why do we want to do
this on the GPU? |
|
Firstly,
this leads to very little CPU utilization. The CPU is used only for
decompression, which I’ll be talking about later. If the data is synthesized
the CPU’s role is very minimal. |
|
Secondly,
there is very little bus utilization. The bus needs to be used only for
upload of decompressed terrain residuals. |
|
We used
shared vertex and index buffers in all levels resulting in a very little
video memory requirement. I’ll be talking more about this later duirng the
talk. |
|
Finally
, we hope to attain much better frame-rates than a similar implementation on
the CPU. |
|
Please
note that using the GPU is possible only because each of the clipmap levels
is a fully regular grid. |
|
Our
implementation uses DirectX 9 and Shader model 3 supported by the nvidia
geforce 6800 card. |