I'm running a compute application (GLSL) on a Radeon HD 7850 (by sapphire with 1GB DDR5 ram)
it runs very slow compared to some Nvidia cards (650Ti), that should be similar in opengl performance,
I've checked with some opengl benchmarks, which confirm that it should be around the same performance
but it's far from it in my application.
The application I'm running does the following:
I'm comparing 2 fixed images to a big matrix of many images, so I have 3 input textures,
two 160x120 and one 5x60 times 160x120 (7200x800). For each image in the big texture
I compare the pixels to the 2 small textures and summarize every column into a new texture.
Everything is done in the fragment shader. The access pattern is such that each thread/shader
invocation summarize a 4 float column (120 pixel height) from the original textures.
I'm sampling one pixel of the large texture once and 5x60 times each pixel of the small textures.
the application is 2-3 times slower then comparable Nvidia cards.
I've downloaded and installed the latest AMD drivers today (not beta).
I'm running on 64 bit windows 7 with a core i7 2600 cpu.