E4 Benchmarks EnginSoft CFD on ARM64

Print Friendly, PDF & Email

RTEmagicC_EnginSoft.tifToday E4 Computer Engineering announced the results of tests carried out independently on a GPU cluster provided to EnginSoft Italy, a premier global consulting firm in the field of Simulation Based Engineering Science (SBES).

E4’s mission is to deliver leading-edge high performing solutions to our wide range of customers,” said Piero Altoè, vice president of research & innovation, E4 Computer Engineering. “Our ARM-based X-Gene clusters coupled with GPUs are already implemented in universities and research centers, but it was a fantastic opportunity for us to have tests carried out by Enginsoft as this firm operates very closely within manufacturing and automotive environments. The results show that we can really make an impact in terms of performance and TCO if we implement accelerated ARMv8 cluster with high speed networks.”

The objective of these tests was to get an independent perspective from a large company working heavily on 3D simulations to determine if the performance and power-efficiency of the X-Gene-based ARM processor is suitable for Computational Fluid Dynamics applications. Each GPU cluster provided by E4 incorporated nodes hosting one GPU, an 8-core, 2.4 GHz 64-bit X-Gene server SoC and a low latency network to enable comparative tests with standard x86-based clusters. The tests provided EnginSoft the opportunity to leverage their experience in CFD and HPC and examine the first production ARM64 processor architecture coupled with a GPU in a real cluster, connected via a FDR InfiniBand, and enabled E4 Computer Engineering and AppliedMicro to have independent tests on the new E4 system.

To test the real capabilities of the cluster in an engineering environment, EnginSoft performed a complete porting of CFD code with a computational engine based on a GPU in order to prove scalability with low power, cost-effective CPUs. The CFD code used for tests was Sailfish, a free computational fluid dynamics solver based on the Lattice Boltzmann method and optimized for modern multi-core systems including GPUs.

e4After an initial software recompilation task, which was supported by E4’s R&D team, Enginsoft succeeded in having the full 3D CFD software solution working with approximately the same performance as competitor’s conventional high-end x86-based platforms and significantly better power efficiency.

For EnginSoft it was an interesting opportunity to be the first company trying to successfully run CFD code on low power ARM andGPU platform,” said Gino Perna, CTO EnginSoft. “The results were very promising in terms of scalability and performance, and there were no doubts that this outcome would have been impossible to achieve only a few months ago.”

The tests results were presented at Teratech forum, on June 23th, in Paris in the paper: Scalability & performances of general purpose fluid dynamics solver on low power cluster: new perspectives on combined CUDA-ARM architecture. EnginSoft is involved in H2020 Exascale projects for testing engineering codes on these new platforms and to improve efficiency on such kind of architecture.

The combination of a high performance, low power, cost-effective X-Gene ARM processor coupled with leading edge GPU compute enables excellent performance at compelling price points,” said John Williams, vice president of marketing at AppliedMicro. “AppliedMicro is excited to see continued validation of the value of the X-Gene processor in E4’s High Performance Computing platforms.”

Sign up for our insideHPC Newsletter