[Nottingham] OpenGL and OpenCL on nVidia GeForce 400 Series

[Mat Booth]

On 9 December 2010 00:57, Martin <martin at ml1.co.uk> wrote:
> Folks,
>
> Anyone know how serious an issue this is?
>
> nVidia GeForce 400 Series OpenGL Problems
> http://en.wikipedia.org/wiki/GeForce_400_Series#OpenGL_Problems
>
> The problem is that the GTX470/GTX480 suffer half or 1/4 the transfer
> speeds for glReadPixels() compared to the previous generation cards
> GTX280/GTX285.
>
>
> Is that a killer for OpenCL calculations? Or just a slowdown for 3d
> render applications such as Blender?
>
> Is it a problem through the entire 400 series range?
>
> Keep with the 200-series? Or does the new Fermi architecture include
> 'must-have' features?
>
> Cheers,
> Martin
>


Interesting link, I hadn't heard about this problem. I guess it just
depends if your OpenCL program does lots of FrameBuffer<-->MainMemory
transfers as that is where the problem seems to be. Reading the
citations however, it looks like you can workaround the problem by
using a pixel buffer object transfer instead, which is a DMA transfer
(as opposed to a CPU mediated transfer).

Unfortunately I don't know if that translates into something you can
do in OpenCL; while occasionally I have cause to get my hands dirty
with OpenGL, I've never used OpenCL before.


-- 
Mat Booth
http://fedoraproject.org/get-fedora