A few cache-sensitive applications (e.g., mcf) lose performance for the 4-core system compared to single-core due to sharing of the L3 and bandwidth resources. Other benchmarks have worse 4-core performance vs. 2-core for the same reason. However, most benchmarks see significant performance improvements (up to 139%). In 21 of the 29 benchmarks, we observe a speedup (with a geometric mean 21.3%) for a 4-core system over a 2-core system. This clearly illustrates the necessity of low-voltage operation to scale multi-core performance, and the need to preserve low-voltage cache capacity to avoid losing the multi-core performance benefits.