I run the NVIDIA CUDA samples v5.0.
In my experiments, GTX 680 doesn’t do faster than GTX 570 for all samples. (ex: stereo disparity computation)
In wiki, GTX 680 has 3090.43 GFLOPs but GTX 570 only has 1405.4 GFLOPs.
- Does anyone know the reasons?
- If I change my platform from Fermi to Kepler, how can I estimate the performance?