Enrolment options

The goal of the course is to provide an overview on the infrastructures specialized for IA with a full-stack approach ranging from hardware to the upper software layers.

First, the course presents the unique characteristics of AI workloads and their specific challenges. It then presents the specific features of modern server architectures addressing compute, memory, and communication aspects, as well as their evolution.

Data center for AI architectures are then presented : physical architectures, protocol stacks, scale out/scale accross concepts and latest technological evolutions are addressed.  

Lastly, the course introduces the principles behind designing and efficiently managing these systems.

The course features tutorials, practical labs, use cases studies and presentations from technical experts (Cisco, Mistral, Nvidia) and a mini-project on Agentic AI.

Keywords: GPU/TPU, GPUDirect, SmartNIC, DPU, Data Center, RDMA, ROCEv2, NCCL, Agentic AI, MCL, Kubeflow, Langchain/LangGraph, KV-Cache, KV-Store.

Guests cannot access this course. Please log in.