Intel multicore processors as platform for research

1
Intel multi-core processors as platform for
research

• Martin Martinov
• Sofia University

2
Overview of the presentation

• Performance Scaling
• Introduction to multi-core processors
• Example from the Life Science Industry
• Conclusions

3
Performance Scaling

• Existing software code runs faster, without
  requiring software engineers to revise or
  optimize their code
• Increasing clock frequencies is no longer viable
  as the primary means for boosting processor
  performance
• New strategies are needed to maintain historic
  rates of performance and price/performance
  improvement

4
Moving to Multi-Core

• In essence, a multi-core processor is similar to
  a multi-processor server, except that the
  parallel compute resources are integrated into a
  single chip
• It also has the potential to deliver better
  performance through faster core-to-core
  communications
• Intel estimates that integrating more cores into
  its processors will lead to performance gains as
  high as 10x within the next three to four years

5
Moving to Multi-Core
Figure 1. The combination of more cores per
processor and thread-optimized software will help
to drive rapid increases in performance and power
efficiency over the next few years
6
Example from the Life Science Industry

• the Life Science Industry has been one of the
  fastest growing worldwide markets for scientific
  computing hardware and software
• It provides a performance analyses based upon
  three commonly used applications in the Life
  Sciences BLAST, ClustalW,(including an MPI
  version), and MrBayes
• These tools exercise diverse computational
  aspects of the hardware sub-system, including
  integer and floating point performance as well as
  small and large sets of input data

7
Example from the Life Science Industry

• Summary Results
• The system performs as an 8 CPU system with no
  significant bottlenecks or flaws in parallel
  scalability
• Hyper-threading improves the scaling of both
  multithreaded and message passing parallel codes
• The Intel Compiler produces binaries which
  outperform those built using the GNU collection
  by 10 to 20 on these applications

8
Conclusions

• A lot of students are coming out of the
  university now having never written a parallel
  program
• We are going to find a significant learning curve
• Intel Software College training in some cases
  helped scientists increase critical application
  processing on by as much as 70 percent
• Faster application processing accelerates the
  pace of research, which maximizes funding and
  speeds scientific discovery

9
Thank you !