As soon as the new generation Intel Xeon processor 5500 ("Nehalem") servers came into production, Intel offered CERN openlab the opportunity to evaluate this new microarchitecture. We chose to evaluate three different flavours of the processor - L5520, E5540 and X5570 – since they provide different levels of performance and power consumption.

Their efficiency was evaluated by measuring the typical power consumption, using standard benchmarks to stress the different subsystems in the server. We also evaluated the performance of the processors with the C++ subset of the SPEC2006 benchmarks. As soon as the new generation Intel Xeon processor 5500 ("Nehalem") servers came into production, Intel offered CERN openlab the opportunity to evaluate this new microarchitecture. We chose to evaluate three different flavours of the processor - L5520, E5540 and X5570 – since they provide different levels of performance and power consumption.

Their efficiency was evaluated by measuring the typical power consumption, using standard benchmarks to stress the different subsystems in the server. We also evaluated the performance of the processors with the C++ subset of the SPEC2006 benchmarks.

According to our measurements, the L5520 is the most efficient overall, being 36% better than the previous Xeon 5400 ("Harpertown") servers, the other 5500 flavours reaching 30%.

Improved efficiency was not the only positive point, since the Nehalem introduces Turbo mode and reintegrates SMT (Simultaneous MultiThreading). In the new Xeon generation, SMT allows each processor to execute two threads simultaneously by sharing the execution pipelines. SMT was thoroughly evaluated, because it may offer a further advantage for a computer centre: increasing the throughput of processed jobs by 15 to 21% based on our tests.

This evaluation involved multiprocessing (using a Monte Carlo based benchmark, "test40") and a multithreaded benchmark ("tbb") based on the ALICE High Level Trigger and the Intel Threading Building Blocks as well as a real-world complex framework (from ALICE) and compared the efficiency of different global scheduling policies.