NECSTSpecial Talk - CUDA Programming Model for Hopper Architecture

Alejandro Chacon
Martin Marciniszyn Mehringer
(NVIDIA)
Event will be online from Facebook
December 14th, 2022
2.30 pm
Contacts:
Marco Santambrogio
Research Line:
System architectures
Martin Marciniszyn Mehringer
(NVIDIA)
Event will be online from Facebook
December 14th, 2022
2.30 pm
Contacts:
Marco Santambrogio
Research Line:
System architectures
Sommario
On December 14th, 2022 at 2.30 pm, a new appointment of NECSTSpecialTalk titled
"CUDA Programming Model for Hopper Architecture" will be held online via Facebook by Martin Marciniszyn Mehringer (NVIDIA) and Alejandro Chacon (NVIDIA).
This session will introduce new features in CUDA for programming Hopper architecture.
The new programming model for Hopper is more hierarchical and asynchronous.
CUDA programming for Hopper introduces optional level of hierarchy called Thread Block Clusters that enable multiple thread blocks within the cluster to communicate using a common pool of shared memory. The asynchronous data movement is now hardware accelerated in all directions between global and shared memories. Additionally, we will cover the new Math API supporting Hopper DPX instructions. We will look at how to exploit the new programming model and APIs in applications for performance tuning.
The NECSTLab is a DEIB laboratory, with different research lines on advanced topics in computing systems: from architectural characteristics, to hardware-software codesign methodologies, to security and dependability issues of complex system architectures.
"CUDA Programming Model for Hopper Architecture" will be held online via Facebook by Martin Marciniszyn Mehringer (NVIDIA) and Alejandro Chacon (NVIDIA).
This session will introduce new features in CUDA for programming Hopper architecture.
The new programming model for Hopper is more hierarchical and asynchronous.
CUDA programming for Hopper introduces optional level of hierarchy called Thread Block Clusters that enable multiple thread blocks within the cluster to communicate using a common pool of shared memory. The asynchronous data movement is now hardware accelerated in all directions between global and shared memories. Additionally, we will cover the new Math API supporting Hopper DPX instructions. We will look at how to exploit the new programming model and APIs in applications for performance tuning.
The NECSTLab is a DEIB laboratory, with different research lines on advanced topics in computing systems: from architectural characteristics, to hardware-software codesign methodologies, to security and dependability issues of complex system architectures.
Biografia
Alejandro Chacon is an AI Developer Technology Engineer (Devtech) at NVIDIA working in the intersection of DL and Bioinformatics. He obtained a MS in computer architecture and information theory from UAB (Barcelona) and a PhD in High Performance Computing for Genomics in the same institution. Consequently, he has been actively contributing to Genomic sequencing projects for HPC over the past 15 years. Before he joined NVIDIA, he held various roles in the industry as computer architect and technical leader in genomics. In his current position, he is responsible for optimizing bioinformatic applications for NVIDIA GPUs.
Martin Marciniszyn Mehringer is an AI Developer Technology Engineer (Devtech) at NVIDIA.
He obtained a MS in computer science from TU Munich in 2002 and a PhD in discrete mathematics from ETH Zurich in 2006. After his graduation, Martin worked in the high frequency trading industry for almost 15 years. Before he joined NVIDIA, he held various roles in quantitative research and development of low latency trading strategies. In his current position, Martin is responsible for optimizing training and inference of deep neural networks with NVIDIA GPUs for financial services.
The event will be held online by Facebook.
Martin Marciniszyn Mehringer is an AI Developer Technology Engineer (Devtech) at NVIDIA.
He obtained a MS in computer science from TU Munich in 2002 and a PhD in discrete mathematics from ETH Zurich in 2006. After his graduation, Martin worked in the high frequency trading industry for almost 15 years. Before he joined NVIDIA, he held various roles in quantitative research and development of low latency trading strategies. In his current position, Martin is responsible for optimizing training and inference of deep neural networks with NVIDIA GPUs for financial services.
The event will be held online by Facebook.