Parallel Systems -- Exercise 1

Course number 67605 --- 2002/2003

Exercise 1: Trends in High Performance Computing

What to do

In this exercise you have a choice of doing either of two things.

Version 1: Scalar vs. Vector Processors

For the last decade or so it has been expected that commodity microprocessors will overtake supercomputer vector processors. Has this finally happened? Well, it depends who you ask and how you measure it.

Moore's law doesn't really give the answer, because it talks of transistors on the chip. But this is highly correlated with performance. Clock speed is also only an indicator, as the question is what you acutally do in each of those cycles.

So, to answer the question, you should collect data about clock rates and MIPS/MFlops ratings of various microprocessors and vector processors, and plot them as a function of the year they were introduced. For starters, you can get information (especially about older processors) from the CPU info center and the survey of high performance computing systems. Use Google to search for more recent data.

So, are things really growing exponentially? Who is leading the race?

Version 2: Trends in the Top500 list

The Top500 list has been published twice a year for 10 years. This allows various trends to be observed, and predictions to be made. An obvious one is that all machines in the list will have TFlop capability by 2005, and that the petaFlop will be reached around 2010.

Can you make other interesting observations? Candidate topics include

10 yearly lists are available in a simple ASCII format at /cs/course/2002/parsys/top500/. You can process them using Perl scripts (or programs in any language you like), load them into Matlab, or whatever.

Submit

Send an email with your results to parsys@cs. Include a description of what you did, and what you found. If you created any graphs, add them as gif/jpg attachements.

Due date: Tuesday, March 11 in the morning, so I can go over them before class on Wednesday. (note extension due to snow last week)