A year ago, the mathematician Cédric Villani delivered his report on artificial intelligence and declared that he wanted to make France a champion of AI at the service of man. To train the necessary talents, IBM has created with its partner Hardis, a SaaS platform dedicated to AI for engineering schools.

Historically at IBM there are two artificial intelligence offerings. The first that everyone knows, Watson, is available in the Cloud and refers to about fifty specialized APIs ready for use. The second, PowerAI also called Watson Machine Learning Accelerator, is a platform aimed at data scientists and developers. In 2018, IBM and its partner Hardis decided to offer PowerAI in hosted mode (Cloud on Power for AI). Grenoble INP – Ensimag (National School of Informatics and Applied Mathematics of Grenoble) was the first engineering school to test and adopt this offer. For IBM and Hardis, the goal is to make the latest hardware and software developments in AI available to all engineering schools and science universities.

A technical architecture designed to optimize AI treatments

Today, more and more engineering schools and universities want to focus on AI to support their students and their teacher-researchers and prepare them for the main challenges of their future professional life, benefiting from the latest innovations in Artificial Intelligence.

Therefore, the machine used in this Cloud on Power for AI offer is a Power9 AC922 server, optimized for artificial intelligence calculations; it is also the base node of the most powerful computer on the planet, Summit, at the top of the TOP500 ranking.

The Power AC922 server delivers unprecedented performance for high-performance computing and artificial intelligence applications.

This server with NVLink technology has Nvidia Volta V100 GPU cards. From a technical point of view, its strong and differentiating point is to interconnect GPUs with each other, but also CPUs and GPUs. Compared to x86 architectures with their traditional connections, the NVLink bus has 5.6 times more bandwidth. The unified view of the server's main memory and the memories located on each of the graphics cards make gradient backpropagation calculations particularly efficient [1]. IBM carried out sizing tests of the offer during the summer of 2018 with Grenoble INP – Ensimag, after which it proved advisable to propose two qualified options of Standard and Premium: the first allows to assign 1 GPU so that a school can handle the practical work of a group of 40 students, the Premium offering sees double the GPUs to enable the processing of work from 80 students connected to the system.

It should be noted that several establishments can group together and therefore subscribe to a single offer of their choice.

The flexibility of Saas mode available to schools

This AI-optimized architecture is available in SaaS mode with a high level of service. It is managed by Hardis, which provides support, support and backup environments. The software offering integrates the most well-known open source frameworks and libraries dedicated to artificial intelligence, such as TensorFlow, PyTorch, Chainer, Anaconda, Keras, etc. Each school has its own dedicated Docker environment. Orchestration of student work is provided by Spectrum LSP software.

Billed monthly based on an annual commitment, these two offerings are designed to enable engineering schools to establish hands-on AI-related work. With Cloud on Power for AI, students move to model what they want. Be it artificial intelligence, machine learning or deep learning.

About Grenoble INP – Ensimag

In the 50s, Jean Kuntzmann, a Grenoblois mathematician, participates in the birth of computer science by developing numerical analysis and mathematics for the engineer, bringing mathematics and physics closer together. From 1962, he will focus on mathematics for computing and logic design. Grenoble INP – Ensimag was founded under his impetus in 1960.

Highly research-oriented, Grenoble INP – Ensimag is the first school to subscribe to Hardis' Cloud on Power for AI program

At the inauguration of the IBM / Grenoble INP – ENSIMAG partnership on March 6, 2019, Catherine Chauvois, Director of Systems IBM France, and Jean-Louis Roch, Director of Grenoble INP – ENSIMAG, officially launched the first treatment of artificial intelligence Grenoble INP.