Evaluation of the Intel Westmere-EX server processor

One year after the arrival of the Intel Xeon 7500 systems ("Nehalem-EX"), CERN openlab is presenting a set of benchmark results obtained when running on the new Xeon E7-4870 Processors, representing the "Westmere-EX" family. A modern 4-socket, 40-core system is confronted with the previous generation of expandable ("EX") platforms, represented by a 4-socket, 32-core Intel Xeon X7560 based system – both being "top of the line" systems. Benchmarking of modern processors is a very complex affair. One has to control (at least) the following features: processor frequency, overclocking via Turbo mode, the number of physical cores in use, the use of logical cores via Symmetric MultiThreading (SMT), the cache sizes available, the configured memory topology, as well as the power configuration if throughput per watt is to be measured. As in previous activities, we have tried to do a good job of comparing like with like. In a "top of the line" comparison based on the HEPSPEC06 benchmark, the "Westmere-EX" platform provides a 39% advantage over the "Nehalem-EX" one. In this comparison the power consumption remained constant, yielding an appreciable 39% of throughput per Watt improvement. Other benchmarks had similar scores, between 14% and 46%, depending on the configuration of SMT, Turbo and frequency scaling. The benefits of SMT remained constant at around 25% for throughput based applications, but were far lower (7%) for an OpenMP based latency bound program. In addition, Turbo mode efficiency was explored and compared in depth for the first time. While before Turbo mode was found to provide large boosts for low active core counts and no boosts for high active core counts, the situation with "Westmere-EX" was the opposite. Turbo mode seemed to provide small but consistent improvements across all active core counts.