Howdy, Stranger!
It looks like you're new here. If you want to get involved, click one of these buttons!
Categories
- All Discussions1,524
- General534
- Graphics109
- GPU Computing419
- Mobile141
- Pro Graphics163
- Tools158
In this Discussion
- Detlef Roettger February 9
- eliyam30 February 12
Tags in this Discussion
- scenix 87
Asynchronous Read-back
-
Hi all,
Is it possiable to do Asynchronous Pixel read from several TextureGL?
Thus writing several textures to host simultaneously
For example: taking snap shots of rotation around an object,
than several FBO's will simultaneously read pixels back to the host (different target files)
target1.jpg .. target10.jpg
(assuming here MRT FBO's) -
2 Comments sorted by
-
Please search for the topic "PBO asynchronous readback" on the web.
While there are plenty of explanations how to optimize the transfer rates that way (there is even a benchmark in the NVIDIA SDK 9.5, standalone download here: http://developer.download.nvidia.com/SDK/9.5/Samples/screenshots/samples/TexturePerformancePBO.html ) if the end result is to compress these to JPGs and store them to disk, your bottleneck will be most likely be the compression algorithm.
Do have a solution for the "multiple views in FBO MRTs" which is faster?
What is your scene statistic? Number of vertices and faces? -
Detlef Roettger said:
if the end result is to compress these to JPGs and store them to disk, your bottleneck will be most likely be the compression algorithm.
Currently , its taking about 3ms to render, 8 ms to read back to host, 10 ms to save to disk (compression)
These "atomic" operations can be divided and done simultaneously.
thus reducing bottleneck.
(4 threads to perform operations separably, should result with ~X4 speedup)Detlef Roettger said:
Do have a solution for the "multiple views in FBO MRTs" which is faster?
I have not tried it yet,
Im just suggesting that theoretically it can be done.Detlef Roettger said:
What is your scene statistic? Number of vertices and faces?
~100K Faces
