Communication on the IBM SP has suffered from the high network latency exacerbated by the software overhead incurred by message passing layers such as MPL. SP AM is a very efficient implementation that reduces communication overhead, delivers high small message throughput, and achieves high bandwidth for bulk data transfers. Split-C benchmarks show that low message overhead and high throughput alleviate the high network latency caused by limitations of the SP network interface architecture. The Split-C and MPI over SP AM results demonstrate that AM is a good substrate for building parallel programming models such as Split-C and MPI without compromising performance. The AM version of Split-C outperforms the MPL version and the MPI NAS benchmarks show that the performance of MPI-AM is comparable to that of MPI-F.