Spark Racing Technology - Further development of a CPU architecture and performance enhancement

Spark Racing Technology – Evolution of CPU architecture and performance boost

Founded by Frédéric Vasseur, Spark Racing Technology is a company specialising in the development and engineering of high performance electric mobility cars and modules. The foundation of the company coincides with the birth of the Formula E project. In November 2012, the promoter and organiser of Formula E, Formula E Holdings, declared that Spark Racing Technology was officially commissioned to design and build the 40 Spark-Renault SRT_01E single-seaters.

Project details.

Services

Professional Services
Software
Training Services
Networking Solutions
Server & Storage

Industry

Manufacturing

Manufacturer

Supermicro, AMD, Bright Cluster Manager, BeeGFS, NICE Software

Company size

< 250 employees

SPARK RACING TECHNOLOGY – ULTIMATE ELECTRIC MOBILITY

THE SHOCKINGLY AMBITIOUS SPARK COMPANY IS FOLLOWING A UNIQUE GOAL – ULTIMATE ELECTRIC MOBILITY

Spark Racing Technology is a contender on the fiercely competitive electric race car market. The company puts its unique capabilities and vast know-how at the service of its customers and partners, enabling them to benefit from state-of-the-art electric mobility solutions for their racing divisions as well as for the general automotive market.

Spark’s headquarters are located in the south of Paris (France). There, the development department, technical planning and construction are combined under one roof. Right nearby, Spark can test its prototypes on all types of terrain. This allows the company to act quickly and optimise quality control and its carbon footprint.

Project objectives.

Spark Racing Technology ran an HPC cluster, but its heterogeneous architecture and age no longer met current user needs.

This cluster was mainly used for the StarCCM+ software, which performs computational fluid dynamics (CFD) simulations.

The need for a new, state-of-the-art cluster, both in terms of CPU performance and parallel storage, was quickly felt. Users wanted correspondingly larger mesh sizes. To optimise the exchange between the computing nodes, the new cluster also needed a very powerful network.

It should also be possible to access the cluster remotely and do graphical post-processing, also via an internet connection for users who want to submit projects from home.

Solution.

Bechtle made all initial strategic decisions and selected the tech partners who would be involved in the implementation of the future HPC infrastructure, choosing Supermicro for the servers, to maintain homogeneity and control the maintenance tools, Bright Cluster Manager to facilitate the overall management of the solution and manage multiple architectures with sub-clustering, BeeGFS for setting up storage for not only powerful but above all scalable computations without questioning the original decisions, and finally, the NICE DCV Suite and EnginFrame to enable graphics sessions on any network.

Together with the teams from Spark Racing Technology, the two-socket architecture on Supermicro AS-2124BT-HNTR with the CPU generation ROME* was selected. This corresponds to a total of 512 simultaneous physical cores (all with a turbo frequency of 3.35 GHz) on a row of 8 computing nodes.

The BeeGFS storage is based on a multi-target server that includes both the metadata on SSD and the data on large mechanical drives. This results in an effective 22 TB and more than 1.1 GB/s total bandwidth!

The installed NICE suite with DCV (high performance remote display protocol) and EnginFrame (HPC resource access portal) allows users to submit and visualise simulations in a GPU-accelerated desktop graphics session from anywhere and from any device.

A hardware and software maintenance contract with Bechtle as SPOC (Single Point of Contact) and dedicated HPC support tokens ensure that Spark receives the necessary support to develop the company further with this strategic tool.

The rollout of this solution took about ten days, including the installation of the hardware in a new rack, the configuration of the software solution and the integration and testing phase of the applications. The Spark teams should receive a turnkey solution and be given all the tools they need to use and manage the cluster as well as possible. Therefore, they were trained and received documentation of all solution components in a final step.

* AMD Epyc Rome – The second generation of AMD Epyc server processors

Business benefits.

The benefits envisaged and desired in the run-up to the project were achieved—easy management through the Bright Cluster Manager, performance and accessibility of the storage, and remote access to cluster resources. Above all, a great deal of time is saved in carrying out the simulation calculations, which was the main objective of the project.

With our new computing cluster, we are recording an average performance gain of 3.5x compared to the old solution. The calculation times correspond to the estimate we had made when working out the solution.

Romain Peretmere, CFD Aerodynamic Engineer