Skip Navigation CECS Home | ANU Home | Search ANU

COMP8320 Tutorial 01 -- week 1, 2011

The Advent of Multicore

Please read the on-line articles below before the tutorial.

• The Free Lunch Is Over
• Serial computing is dead; the future is parallelism (handout - the link now reqires a login):
• State-of-the-art for parallel compilers
• The Case for CMP, Olokuton et al (handout)

Discuss the following questions:

1. What is meant by the term `free lunch'? Specifically, what architectural features are involved?
2. What trends related to Moore'e law have stopped over the last decade (and why?) and what are continuing?
3. Why is concurrent (parallel) programming difficult? Give specific scenarios to illustrate this.
4. What are the current applications that are well-suited to multicore?
5. Is the learning of a parallel language sufficient to meet the challenge of writing software for multicore? Discuss from several perspectives.
6. What to factors have resulted in the serial (von Neumann) programming model persisting for so long?
7. Clock speed increase has occurred naturally as a result of Moore's Law (reduction of transistor size). What is the other factor which has contributed?
8. The three main architectural techniques for increasing Instruction Level Parallelism are becoming increasing expensive and/or cost ineffective. What are the reasons why in each case? What is the relationship between the cost and the potential gains in ILP? What meanings does the word `expensive' in any case?
9. Why might a 6-issue processor have larger pipeline stalls than a 2-issue (with comparable clock speed)?
10. Critically assess the main conclusion of The Case for CMP, that a CMP made of simpler cores is 30-100% better than a single core chip trying to extract maximum ILP. What assumptions are in this? What qualifications should be given?

Copyright | Disclaimer | Privacy | Contact ANU