I am using open mpi to run a parallel code. I am using an 8 core AMD processor. The code that i have made has a limitation that it can be run only on odd no. of processors. So when i run the code on 7 processors, all the cores show 100% usage (i.e. total cpu usage=87.5%). But when i run 2 codes using 5 and 3 cores respectively, the performance falls to 50% of 6 of the cores i.e. (total cpu usage =63.5%). Can someone suggest how to improve the performance of the cores?

It's difficult to help without knowing more about your code. It's impossible to use 100% of all 8 cores as the operating system does need a little bit of CPU for things like interrupt handling, background procesees, and the like. Have you run any timing benchmarks of your code to see that performance is really worse than expected in the 5+3 case? Are you sure that the code wasn't just doing heavy I/O (which would drive down CPU usage) when you looked? Running two instances of the program could cause a much heavier I/O load which will slow down overall performance.

It's possible that your application is simply highly sensitive to OS jitter or is heavily I/O dependent (so having two copies running will stress your machine's I/O system beyond its capabilities). In that case, you need to find the bottleneck, and it may be difficult to have two instances of your program running at once without working around said bottleneck.