Here you can see a comparison of our encoder in GPU and CPU version with several other JPEG2000 CPU implementations, both commercial (Kakadu) and open-source (JJ2000, OpenJPEG).
The test images can be found here (in the RGB-8bit package). Our test system:
Our GPU version on 280GTX is faster than all of the single-thread-CPU encoders for images larger than 33 MB, for both reversible and irreversible encoding. The speedup to Kakadu running as single thread with 83 MB input data is 2.6x for reversible and 3.1x for irreversible encoding.
The multithread-version of Kakadu is faster until the 83MB input in lossless mode, in lossy mode it is still as fast as our encoder for the 83MB input.
For more detailed benchmarking, see our documentation.