I'm having a difficult time understanding the difference between the linear algebra packages MAGMA and PLASMA from just a quick glance. It looks like MAGMA is oriented towards GPU's and vector coprocessors, while PLASMA is oriented towards a vanilla multicore system. But I'd appreciate an explanation comparing them, and what they do differently from BLAS/LAPACK/ScaLAPACK.
See here or the PLASMA README. LAPACK is serial and parallelism enters only via multi threaded BLAS3. ScaLAPACK is for distributed memory machines and requires BLACS/MPI. PLASMA targets shared memory parallelism on multicore machines.