|
|
|
EDUCATION
|
|
Heterogeneous Computing
A new class, Heterogeneous
Computing,
was offered in the Fall semester in National Tsing Hua University. The
topics covered in the class can be roughly divided into four groups:
- CUDA basic: CUDA syntax, CUDA memory, GPU architecture.
- CUDA optimization: CUDA stream, techniques for
performance optimization, CUDA profiler and debugger, performance
libraries
- Multi-GPU programming: Multi-thread programming and MPI
programming.
- Others: OpenACC, OpenCL, other accelerators, Kepler and
CUDA 5
Another important part of the class is the term project, for which
students can team up to find an interested topic to study and to
implement it on GPU. Through the project, students demonstrate their
ability in CUDA programming and performance tuning on new problems.
|
|
CUDA Programming Contest
Nvidia Taiwan, National Center of
High-performance
Computing (NCHC), and CCOE
co-hosted the third CUDA
programming contest in Taiwan to promote the novel usage of CUDA in
researches or works. The theme of this year is "Big Data", which is one
of the major challenges in many areas. The contest has two stages. In
the first stage, contestants submit their proposals; and in the second
stage, the selected contestants present and demonstrate their CUDA
programs on a GPU cluster, Formosa
5, which is provided by NCHC.
|
|
SCC Training
The student cluster competition (SCC)
is one of the student events held in conjunction with the
Supercomputing Conference (SC) since 2007. The competition is a two
days non-stop marathon for students to run HPL, HPCC, and four high
performance computing (HPC) applications on a small cluster assembled
by each team. The only limitation of the machine is the total power
consumption. From 2010, some HPC applications have CUDA version, such
as QMCPack. Since then, the CUDA programming and performace tuning
become part of the SCC training.
Some of the video for 2012's final (1) (2).
|