Edge System and Use Cases

Emerging applications stream

EPI’s ambitious goal is to set an architecture suitable for different applications. In SGA2, new emerging applications will be studied. In SGA2, a new emerging applications will be studied: Autonomous HPC with use cases deployed on the edge. But this application is considered in continuum computing from Autonomous HPC on edge to performant HPC nodes in data center.

Autonomous HPC

The main objective is to demonstrate the capability of EPI architecture for Autonomous HPC use cases. This demonstration will be done first on one of the previously selected use case: the Video Surveillance for smart city and infrastructures.

The goal of these explorations is to find and optimize the HPC scaling path to fit with the compelling constraints of the (high volume) embedded market, by evaluating, analysing and profiling these use cases against the target HPC processors. This work will be done on the simulation tool chains of all the platforms developed in SGA2. The chip architecture’s requirements will be achieved through expertise for the extraction of the key algorithms from the target application, including security, energy efficiency and ease of deployment requirements. The architectural exploration, based on the kernel computation requirements and the configurable parameters provided by the simulation tools will push through the best “scale-to-fit” of the HPC processors and RISC-V accelerator down to the embedded solutions. The vector and ML accelerators developed in Stream 3 will also be evaluated on real time video surveillance application, to propose extensions to meet the real time constraints with increasing data size.

Moreover, software libraries will be developed to support federated/distributed training and inference of DNNs and feature extraction algorithms for video-based applications for massive surveillance. As such application requires a full scale HPC processor from edge up to the central HPC cloud, co-design with the other streams will define and properly tune all the parameters required.

EPI-Based HPC Blades

For continuum computing, the applications on Edge must be able to exchange with HPC servers. Then the EPI-based blade developed for HPC must provide the adequate network interface for connection to edge network.

HPC Vendors and HPC end-users will analyse how EPI technologies fit in the HPC market and propose valuable HPC products based on General Purpose processor Rhea2 or accelerators developed in SGA2. The different proposals must meet standard HPC requirements, with a focus on security, but also new requirements such as training phase in machine learning for emerging application.

Several scenarios will be analysed for high-speed interconnect between HPC nodes, as interconnect is essential to meet expected performances of HPC Product. The best technology and topology will be selected from the needed bandwidth and latency. The connection of the HPC nodes to the rest of the datacentre and to the autonomous HPC servers will also be carefully analysed.

The next step is, for few nodes defined in the first step, to study the feasibility to implement them in a blade in one standard HPC form factor. This form factor is the OpenSequana one, that is compatible with the BullSequana XH3000 cabinet from Atos.

One type of node will be selected according to the interest of the SGA2 partners and to the feasibility study, and a pre-study of this blade will start using the Rhea2 reference design developed in Stream 2, with a focus on a new water-cooled heatsink for Rhea2, compatible with the OpenSequana interface.



Our website uses cookies to give you the most optimal experience online by: measuring our audience, understanding how our webpages are viewed and improving consequently the way our website works, providing you with relevant and personalized marketing content. You have full control over what you want to activate. You can accept the cookies by clicking on the “Accept all cookies” button or customize your choices by selecting the cookies you want to activate. You can also decline all cookies by clicking on the “Decline all cookies” button. Please find more information on our use of cookies and how to withdraw at any time your consent on our privacy policy.
Accept all cookies
Decline all cookies
Privacy Policy