CUDA Multi-GPU Question

I have four Tesla2050’s cards in the C410x, the use of these cards for general purpose computing,
when I use when developing cuda encountered the following problem:

Development environment as follows:
rhel5.5 operating system, cuda3.2 toolkit, four graphics tesla 2050;
The use of multiple graphics performance is equivalent to a performance?
Is to use two or more pieces of computing performance, computing performance is always equal to one.