Evaluation of the Intel Westmere-EP server processor

In this paper we report on a set of benchmark results recently obtained by CERN openlab when comparing the 6-core "Westmere-EP" processor with Intel's previous generation of the same microarchitecture, the "Nehalem-EP". The former is produced in a new 32nm process, the latter in 45nm. Both platforms are dual-socket servers. Multiple benchmarks were used to get a good understanding of the performance of the new processor. We used both industry-standard benchmarks, such as SPEC2006, and specific High Energy Physics benchmarks, representing both simulation of physics detectors and data analysis of physics events. Before summarizing the results we must stress the fact that 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 Simultaneous Multi-Threading (SMT), the cache sizes available, the memory configuration installed, as well as the power configuration if throughput per watt is to be measured. We have tried to do a good job of comparing like with like. In summary, we see good scaling with the core count. We observed a very appreciable throughput increase of up to 61% when using the in-house benchmarks, compared to the previous processor generation. Our variant of the SPEC benchmark rate, "HEPSPEC06", gives 32% more throughput. HEPSPEC per w att is measured to improve by up to 23% which is less than the improvement when going from Harpertown to Nehalem (36%). Benefits of SMT were seen to be of similar significance as in the previous processor generation.