Credit on the CPU apps is slightly above average vs. the BOINC standard. That being said, while the CPU apps work well, due to the ability to run 16 -800 iterations in parallel on an nVidia CUDA or ATI CAL/Brook GPUS vs. a CPU far outweighs the processing power of a CPU on this project, so machines with CUDA and RC600+ ATI machines will get far more credit per hour.
As did the original 3x+1 app , the CPU apps currently uses the GNU GMP library on the CPU apps whereas the GPU apps use a 240 bit library. At some point, the will cease to return valid results whereas the CPU apps will continue to work. So, while the GPU apps are limited to 2^240 bites, the CPU is virtually unlimited.
In other words, for now, the CUDA and ATI cards will do much, much better on this project than the CPU only versions, even with the CPU SSE or SSE2 optimized apps. Then again, the GPU apps require much more electricity. Got any suggestions as to what to tell my wife why our electric bill is almost double that of last year while we have the same number of computers? J
Jon Sonntag