about SFU in GTX8800

It’s said that GTX 8800 has two SFUs for complex mathemetical computation. Does that mean instructions like __fdividef, __expf will be executed on those SFUs in parallel to other processors, such that if very few complex mathemetical computations exist in the program, they should not impact the overall speed?


I believe that is not the case. I believe it is like double precision. If the SFU’s (or DP unit) are busy, the SP’s are resting.