Reading from Video Card
Author
I am working on a project where we are discussing using the GPU to do High Definition video processing. The question then becomes, what is the best way to get the edited video stream from the video card back to the hard drive? What would be the implications of using the GPU to do video processing?
Cheers,
Bob
Have you any idea what sort of hardware you're using?
It's important as AGP tends to be very fast writing yet relatively slow when reading. From my understanding PCI-E is a lot better and offers the same (?) speeds for both read and write.
I don't know so much about video processing, but I have seen it with some other similar graphics algorithms (where the work is offloaded to the GPU) that they keep a chain of 10 (or more) render targets that store the results for a delay of upto 10 frames. That way you need to only read back 1/10th of a frames data in each cycle. This will only work if you've got a chunk of VRAM spare to use though [smile]
hth
Jack
It's important as AGP tends to be very fast writing yet relatively slow when reading. From my understanding PCI-E is a lot better and offers the same (?) speeds for both read and write.
I don't know so much about video processing, but I have seen it with some other similar graphics algorithms (where the work is offloaded to the GPU) that they keep a chain of 10 (or more) render targets that store the results for a delay of upto 10 frames. That way you need to only read back 1/10th of a frames data in each cycle. This will only work if you've got a chunk of VRAM spare to use though [smile]
hth
Jack
Not especially a pretty technique, but it highly depends upon the API you are using, but if you are running AGP it isn't even worth your while, however if you can wait for DX10(or whatever it is called) I think one of the main features is fast readbacks.
Quote:Original post by Arelius
however if you can wait for DX10(or whatever it is called) I think one of the main features is fast readbacks.
With a PCI-e card...
I think the thing with DX10 is that readbacks can also be cached, which previously they can't. When you lock a surface, you force everything about that surface to be computed, which it may not be at the time of the lock. That causes your app to stall. With DX10, you can readback into another buffer without actually stalling, and then access that buffer later.
This topic is closed to new replies.
Advertisement
Popular Topics
Advertisement