Tip:
Highlight text to annotate it
X
100 Teraflops peak performance
is readily available even for small sized companies and organizations.
Time consuming algorithms are needed in a wide variety of business fields:
Finance Companies want to simulate their liquidity trend within minutes instead of days.
Business intelligence users want more sophisticated algorithms for real time inspection.
Satellite data based retrievals want to use the most accurate algorithm to model their observations.
All this is already possible!
The key to success is to utilize all processing resources at nearly 100%.
This requires changing the algorithms to incorporate fine grained parallelism strategies.
However, those modifications often affect the entire architecture,
and thus many thousand lines of code.
At the high performance computing department at Catalysts
we worked on many projects in high performance computing.
During the last years we refined our way of working.
We are now able to exchange architectural building blocks in a program
within seconds!
We use our automatic and semi-automatic tools
to optimize for a highly parallel, scale-free design.
These tools can detect potential parallelism in a program,
simulate parallelism on a distinct point in the program,
refactor codes to be prepared for higher performance,
optimize the memory access patterns,
generate readable code for special high performance hardware and clusters.
A workflow using this toolchain allows rapidly tweaking an algorithm
for the best possible performance!
We are Catalysts. Software is our Passion.