Using Intel Software tools for HPC and AI in Azure

During the first week of July, Grey Matter had the pleasure of running two workshops with Intel Software, their partner Bayncore, and Microsoft. These workshops took place in Dublin at the Microsoft Offices and in London at the Microsoft Reactor space. The focus was around using the Software tools for HPC and AI in Azure.

Bayncore’s Stephen Blair-Chappell and George Zitzlsberger guided us through a series of hands-on cloud-based labs. Both being former Intel consultants, they brought with them a wealth of experience to the workshops.

The day was split up into two parts – the morning focusing on High Performance Computing on the Cloud, the afternoon looking  at Intel’s solutions for Deep Learning and Machine Learning.

In the morning, first session “Performance optimization of a cloud-based application using the Intel® Compiler”, we took a standard nbody physics code and showed how to optimise the code so it runs well in a cloud computing environment. Using the latest Intel processors on Azure F4_v2 instances (aka Skylake), the attendees ran through a series of optimisation steps that eventually provided a  ten-times performance boost of the application. One of the key contributions to the performance boost was using the Intel compiler to maximise the vectorisation and memory access efficiency of the code.

We then followed up with a cloud based session, “Performance optimization using Intel® Advisor”, in which the attendees used the roof-line analysis from Intel Advisor to plot the improving vectorisation efficient of the same nbody physics code as various optimisations were applied. The roofline model is a great way to show how both the memory read-writes and vectorisation efficiency is impacting on the running of the application.

In the afternoon, in the session, “Faster Machine Learning and Deep Learning training with optimized Frameworks”, we were led through an AI example that used an Intel optimised version of TensorFlow. We learnt how the Intel optimized frameworks can be used for Deep Learning training on Intel CPU based platforms. 

This was followed by a short discussion and presentation on how to deploy trained models on multiple types of targets using the Intel® OpenVINO™ Toolkit.

Here’s what some of the attendees had to say about the event:

“It was a really good and successful event. The level and pace of the material covered was very appropriate and very professionally delivered. Only suggestion would be doing what you’re doing! And thank you very much to Grey Matter for organising the workshop, to Georg and Stephen for presenting it, and to all involved.”… ” I really loved the event and would definitely like to attend more and enlighten my knowledge in the field of technology while sharpening my skills.”

Some wanted more, so much so they wished we could have run it over two days and have more hands-on…”[to improve, have a] 2-day course, with more hands on“… “It should be more interactive”

View the slides and corresponding documentation here.

Let us know how you get on.

We run events regularly with Intel and Microsoft, sign up to our newsletter to be sure to hear about our upcoming technical events.

Grey Matter is proud to be a Microsoft Gold Partner and a Cloud Solution Provider for direct and indirect sales, as well as an Intel Software Elite Reseller.

Grey Matter and Intel are co-hosting another event 10-11 December. Day 1 focuses on HPC software development for Financial Services, and day 2 is about their new model, Intel oneAPI. Find out more and sign up to each of these workshops here.