Leader en solutions logicielles pour scientifiques
Don’t miss the opportunity to dive deep into the latest tips and tools from Intel® Software and learn more about Intel’s resources and hardware. Test them out yourself and consult our experts onsite at this year’s hands-on training for High Performance Computing & Artificial Intelligence!
Join us for 2 days of hands-on coding sessions on Parallel Programming, Performance Optimization, Artificial Intelligence, and Machine & Deep Learning.
Explore exciting topics including scientific and technical computing, computer vision, image and pattern recognition, machine learning, optimized deep learning and big data analytics.
Please bring your own – Intel®-based – laptop1. We will provide all required software and technology. Detailed technical requirements will be sent to registered attendees.
You are welcome to attend both days but you can also attend one day only. Please specify at registration! Attendance is free, please register as soon as possible as this time we only have very limited seats. After your registration we will review and will confirm your attendance as soon as possible!
When it comes to improve application performance, one needs to re-architect and/or tuning existing code to expose enough vectorization and parallelism. In this workshop we will dive into the code modernization framework. This results in a systematic approach which needs to be followed to achieve the highest performance possible. With the help of examples, use cases and a better usage of the Intel® C/C++ compiler, we pinpoint you to possible inefficiencies both on sequential and vectorized code and we explain remedies, hints and strategies to be considered to ensure an application delivers great performance on today’s scalable hardware and upcoming future generations.
AGENDA DAY 1**
This session will introduce intel tools and the different suites available for writing codes for single or multi-nodes computers as well as analyzing the performance.
This session will drive the user into many compilers based optimizations. Step by step, the attendee will be able to understand how to modify the code and the compiler arguments to achieve a great performance.
This session will drive the user into many compiler based optimizations. Step by step, the attendee will be able to understand how to modify the code and the compiler argumuments to achieve a great performance.
Modern architectures offers many way to greatly speedup your applications. Parallelism is one of them. On a single node point of view, parallelization can be achieved by threading and vectorization. This presentation will explain how to create threaded and vectorized workload with OpenMP.
Moving data in an efficiçent way is a crictical point when it is about HPC. Many HPC applications tends to be memory bounded and it is alway a good practice to verify that memory accesses are hadware frendly and that we reuse as much data from the cache as it is possible.
Intel MKL provides many mathematic functions implemented by the most talented engineers at Intel. This library achieves incredible performance. This track will explain how to compile and link with the MKL. We will also show some speedup that can be obtained by using this library.
During the morning session we will show how performance analysis tools like Intel® Advisor and Intel® VTune™ Amplifier can be used efficiently to investigate issues and guarantee the optimal performance for the underlying latest Intel® Xeon® Scalable processor. During the afternoon we will show how to use the Intel® Distribution for Python giving insights of the most used algorithms for machine learning applications and how libraries such as Scikit-Learn has been optimized for the Intel® hardware.
AGENDA DAY 2**
Intel Advisor is a powerful tool for tracking down and solving vectorization problems. This presentation will introduce Intel Advisor and especially the survey and the trip count analyses. We will explain how to read Advisor's outputs to improve the vectorization.
A Roofline chart is a visual representation of application performance in relation to hardware limitations, including memory bandwidth and computational peaks.
This session will drive the user from an unoptimized version of a wave propagation kernel to a much more optimized version. We will see on a real world example how to detect bottlenecks and how to optimize them.
Lloyd's algorithm is the standard algorithm used for K-means clustering, an unsupervised machine learning technique. We'll use it to reduce the number of colors used in an image. First, we'll use Scikit-Learn to perform this task and we'll try to understand why the implementation is so slow. Then we'll move to the DAAL to get better performance. Finally, we'll try to write and improve the performance of our own implementation using Cython.
In this lab, we'll show how Intel has incorporated TBB into the Python ecosystem. We'll first study an example on collaborative filtering, a technique used by recommender systems. We'll show how oversubscription can reduce the performance of the original program and how this problem has been fixed by Intel TBB. We'll then show how Dask can be used on top of TBB for composed parallelism.
