Abstract: The HPC world has been obsessively focused on simple performance metrics, such a s peak floating point performance, or performance on the Linpack benchmark. Howe ver, two critical problems, the unbalanced architecture and the deficient progra mming support, have never obtained a good solution. The recent DARPA High-Produc tivity Computing Systems (HPCS) initiative has raised the issue of proper produc tivity measurements, in the context of high-performance computing. The backgroun d, objective, challenge, and main efforts toward the research of HPCS are presen ted here.