Question about SLI

I’m plan to using NVIDIA’s SLI technology to optimize my program that running in SLI platform(using

SFR rendering mode),I have below questions,and doubt SLI is suitable for me:

  • My program upload some texture’s image every frame,because the image is from movie file or

camera,so it is only valid in current frame rendering,but I heard SLI will copy the image to all

display cards for ensure contents is identical in all cards,is it true?Is it means the image will be

transfered from CPU to GPU twice?

  • My program has some image process in GPU,I use 2 textures and a FBO to do it.My image process has

many steps,so I use one texture as src,the other is dest,after complete a step,swap them,then do

next.It is a method called “PingPong Rendering”,and I like it.But from NVIDIA’s doc,it seems to say I

had better clear the color buffer before every frame rendering so the GPU knowns the RenderTarget

need not to be synchronized.It is not a good news for me,because my image process has many steps!If I

folow the doc’s suggestion,I’ll clear the buffer for many times.So what I understand is right?

  • If I not using SLI,write my own code to distribute rendering job to every GPU,can I get identical

performance than SLI?Or somewhat higher?

Sorry,there is a error,I’m plan to using SLI’s AFR rendering mode

SLI isn’t well-suited for GPGPU applications such as Pingpong rendering.

There’s some discussion on this here:
http://www.gpgpu.org/forums/viewtopic.php?t=3875

You can write code to target each GPU manually, although the methods to do so are somewhat different on Windows and Linux, and are something of a pain on the latter.

Originally posted by pango:
Sorry,there is a error,I’m plan to using SLI’s AFR rendering mode
I hope you don’t plan to use Windows Vista with SLI because…

http://forums.slizone.com/index.php?showtopic=5585
http://support.microsoft.com/kb/936710
http://uk.theinquirer.net/?article=41312

etc etc…