Massively parallel computing
GPUs
Parallel array reduction
Conclusion
Speeding up computations: bring the GPUs Supercomputing for the masses
Florent Bouchez
[email protected] Indian Institute of Science Bangalore
Vignan University, Guntur, November 20th 2009
1/7 F. Bouchez (IISc)
HPC and GPUs
20/11/2009
Massively parallel computing
GPUs
Parallel array reduction
Conclusion
Where is the supercomputer?
My department at IISc: SERC.
2/7 F. Bouchez (IISc)
HPC and GPUs
20/11/2009
Massively parallel computing
GPUs
Parallel array reduction
Conclusion
Where is the supercomputer?
My department at IISc: SERC.
SERC à Supercomputer Education and Research Centre.
2/7 F. Bouchez (IISc)
HPC and GPUs
20/11/2009
Massively parallel computing
GPUs
Parallel array reduction
Conclusion
Where is the supercomputer?
My department at IISc: SERC.
SERC à Supercomputer Education and Research Centre. But I cannot talk to you about supercomputers. . . Blame the Americans for that!
2/7 F. Bouchez (IISc)
HPC and GPUs
20/11/2009
Massively parallel computing
GPUs
Parallel array reduction
Conclusion
Where is the supercomputer?
My department at IISc: SERC.
SERC à Supercomputer Education and Research Centre. But I cannot talk to you about supercomputers. . . Blame the Americans for that! Instead, I can talk about massively parallel architectures: supercomputing for the masses.
2/7 F. Bouchez (IISc)
HPC and GPUs
20/11/2009
Massively parallel computing
GPUs
Parallel array reduction
Conclusion
Outline
Massively parallel computing
GPUs
Parallel array reduction
3/7 F. Bouchez (IISc)
HPC and GPUs
20/11/2009
Massively parallel computing
GPUs
Parallel array reduction
Conclusion
Massively parallel computing
Not everybody has access to a supercomputer. But access to many computers is easier. What are the main differences? De-centralized data. Communication costs. Need massively parallel problems. (e.g. Seti@home)
4/7 F. Bouchez (IISc)
HPC and GPUs
20/11/2009
Massively parallel computing
GPUs
Parallel array reduction
Conclusion
GPUs
GPU Graphic Processing Unit. This is where we jump to the Post-Doc presentation of Sept. 12th.
5/7 F. Bouchez (IISc)
HPC and GPUs
20/11/2009
Massively parallel computing
GPUs
Parallel array reduction
Conclusion
Parallel array reduction
This is where we jump to Mark Harris presentation from NVIDIA. The goal is to show how parallel computing can be powerful, as well as difficult to get right. . .
6/7 F. Bouchez (IISc)
HPC and GPUs
20/11/2009
Massively parallel computing
GPUs
Parallel array reduction
Conclusion
That’s all for today
7/7 F. Bouchez (IISc)
HPC and GPUs
20/11/2009