|
Abstract:
As
computing demands advance to meet the ever increasing needs to solve more
difficult problems, we simultaneously have seen new constraints emerge to
limit how chip manufacturers can meet these performance demands. Today
power and heat constraints have driven chip manufacturers
to using slower clocks and multiple cores. This approach has been enabled
by improved manufacturing and material design resulting in ever dense
chips. This also means that the commodity chips on which our advanced
computing has been built are becoming more specialized. Multi-core chips
from AMD and Intel will force increased emphasis on algorithm development
and implementations that must be tremendously concerned with data location
and computer memory access and latency costs. This creates a catalyst for
a new way to look and develop advanced computing, in particular what Cray
is calling Adaptive Supercomputing. Cray’s adaptive supercomputing is
focused on using heterogeneous computing capabilities in a hardware and
software integrated environment so that the focus of analysis and
solutions is again on the solutions, not on the hardware environment in
which a solution will be computed. In this talk I will trace some of the
technical developments that has brought us here as well as some of the
technologies and computing capabilities which will bring us adaptive
supercomputing and what it will mean.
|