GCOE serves as a hub for all heterogeneous high performance computing activities in the country.
GCOE plays a major role in promoting GPU computing through infrastructure development, manpower training, student internships and projects, and research funding. To increase awareness in GPU computing, GCOE conducts workshops across the country.
This one-day workshop on “parallel computing with GPUs” introduces some of the key tools for parallel scientific computing and embedded computing. The workshop will include lecture-cum-demos and hands-on sessions.
The workshop can be basically divided into three sessions:
A. Parallel computing with MATLAB
B. GPU computing with CUDA
C. Embedded Super Computing with Jetson Tk1 board
A. Session on Parallel Computing with MATLAB:
We look at the concepts of vectorization, parallel MATLAB computing on multiple CPU cores, and parallel
MATLAB computing on GPUs. Participants will be taught how to parallelize loops, distribute or co‐distribute
large arrays over multiple CPUs, and speed up matrix computations on both CPUs and GPUs. All concepts will
be demonstrated through hands‐on sessions.
B. Session on CUDA and OpenACC:
This part will focus on enabling users to write CUDA codes. This will be followed by an introduction to OpenACC
‐‐ a standard for parallelizing existing serial code. Participants will be introduced to the GPU concepts using
single‐block and multi‐block vector additions. Shared memory examples, device query and error handling will
be covered through demos and hands‐on.
C. Session on Embedded Supercomputing:
This part will cover features and applications of embedded supercomputing with NVIDIA’s high‐end Embedded
platform Jetson TK‐1. Jetson TK‐1 gives high‐performance with low‐energy consumption for deep learning and
computer vision applications. Participants will be introduced to hardware and software development on the
Jetson TK‐1 platform, enabling them to explore platform in their project research work.