HPC: Key Technologies and Tools

The NVIDIA AI Technology Center (NVAITC) is a program to enable and accelerate AI research projects in Italy, focusing expertise and resources on specific research projects. The program is a national-level collaboration centred around project-based collaborations with institutions within the CINI network and fosters the collaboration of the Italian Computing Facility, CINECA. It aims at enabling academic institutions at all levels to conduct their research more efficiently by collaborating into research projects, training students, nurturing startups and spreading adoption of the latest AI technology throughout Italy. Example areas of contribution include:

 

·       Adoption of DL/ML frameworks (NVIDIA heavily contributes to DL frameworks development).

 

·       Technology selection and optimization (efficient data loading, mixed-precision, inference).

 

·       Model architectural choices.

 

·       Contribution to software development.

 

·       Performance optimization and tuning through profiling.

 

·       Workload scaling on multi GPUs/nodes.

 

·       Discussion on research studies.

 

·       Training.

 

·       Support to access HPC resources.

 

 

 

In order to participate to the program and receive support, interested PIs can submit a proposal using this template [1] via email. The local engineers (Giuseppe Fiameni and Andrea Pilzer) can be contacted for any input, suggestion, or advice before submitting it. PI is typically contacted back for clarification and SoW settlement after submission. Proposals are reviewed with the help of fellow NVAITC engineers on a first-come-first-serve basis. The review takes a couple of weeks.

 

Evaluation is based on NVAITC criteria (target publication, technology stack and computing scale), rules of engagement (compact timeline, no compute, no funding, etc) and shared realistic expectations (an agreed-upon SoW). This “call for proposals” remains open as long as the program has the capacity to handle projects.

 

Conversely, access to computational resources is handled separately by CINECA via the ISCRA/PRACE programs (iscra link, prace link).


 

Scientific Advisory Board

 

Daniele Nardi

UNIROMA1

nardi@diag.uniroma1.it

Carlo Sansone

UNINA

carlosan@unina.it

Giovanni Farinella

UNICT

gfarinella@dmi.unict.it

Marco Ferretti

UNIPV

marco.ferretti@unipv.it

Marco Bertini

UNIFI

marco.bertini@unifi.it

Tatiana Tommasi

POLITO

tatiana.tommasi@polito.it

Paolo Cignoni

ISTI-CNR

paolo.cignoni@isti.cnr.it

Frédéric Parienté

NVIDIA

fpariente@nvidia.com

Luca Oliva

NVIDIA

loliva@nvidia.com

Sergio Orlandini

CINECA

s.orlandini@cineca.it

 

[1] NVAITC EMEA Project Proposal Template
NVAITC Presentation Slides

 

AI Webinar Series on Deep Learning for CINI AIIS Labs - June 29th/July 3rd 2020
The goal of this webinar series is to explore the fundamentals of deep learning by building and training neural networks, optimizing data loading and performance through mixed-precision and parallelization, and deploying your trained model in production for inference. You will learn how to design, train, optimize, profile and deploy a deep neural network using NVIDIA technologies. Each session is split.  

 

Date

Topic and slides

 

Session 0

Linear Regression in Pytorch - Christian Hundt

Video

Session 1

Convolutional Neural Networks (from slide 45) - Christian Hundt

Video

Session 2

Efficient Data Loading using DALI - Giuseppe Fiameni

Video

Session 3

Mixed Precision Training using Apex - Paul Graham

Video

Session 4

Multi-GPU Training using Horovod - Gunter Roeth

Video

Session 5

Deploying Models with TensorRT - Niki Loppi

Video

Session 6

Profiling with NVTX - Giuseppe Fiameni

Video

Share This

S5 Box

Cini Single Sign ON

Questo sito memorizza solo cookie tecnico/funzionali. Se vuoi saperne di più vai alla sezione Cookie Policy