{{{#!div class="box"
= [/wiki/CSC CSC] NVCUDA =

This [/wiki/CSC CSC] module takes advantage of nvidia GPU's [http://www.nvidia.com/object/cuda_home_new.html CUDA] capability to parallelize the conversion process and speed things up.
You may also consider using the [/wiki/CSC/OpenCL OpenCL CSC] module, which can also use the GPU and is more generic (not vendor specific), better tested and often [/wiki/CSC/Performance faster].
}}}

{{{#!div class="box"
== {{{csc_nvcuda}}} Installation ==
* you need to build xpra with the flag "{{{--with-csc_nvcuda}}}", since this module is not enabled by default.
* you must use a sufficiently recent version of the {{{nvidia}}} video drivers, preferably obtained directly from nvidia, and not the {{{nouveau}}} drivers (which you should probably remove).
* the build file relies on the presence of a standard {{{pkgconfig}}} file to find the CUDA libraries and headers. nvidia does not provide one with the SDK, so you will need to create one. You can use this example [/attachment/wiki/CSC/NVCUDA/cuda.pc cuda.pc] as a template.
}}}

{{{#!div class="box"
== Links ==
* this module relies on [http://documen.tician.de/pycuda/ PyCUDA]
* [https://developer.nvidia.com/cuda-gpus supported GPUs]
* original feature request ticket with more information: #384
* [/browser/xpra/trunk/src/xpra/codecs/csc_nvcuda csc_nvcuda module source]
}}}

{{{#!div class="box"
== Module Options ==
* choose the device used with:
{{{
XPRA_CUDA_DEVICE=N xpra ...
}}}
Where `N` is the device ID.

* Use:
{{{
xpra -d cuda start ...
}}}
 to get more information. (versions older than v0.12 required the use of the {{{XPRA_CUDA_DEBUG=1}}} environment variable)
}}}