Breaking News
News & Analysis

Berkeley kicks off parallel computing lab

3/18/2008 06:00 PM EDT
1 Comment
Page 1 / 2 Next >
More Related Links
View Comments: Newest First | Oldest First | Threaded View
User Rank
re: Berkeley kicks off parallel computing lab
Mapou   3/18/2008 8:04:47 PM
In my opinion, the idea of building a software abstraction layer in order to hide the complexity of the multicore processor from the programmer is pure folly. On the contrary, it is the multicore architecture that should be changed to address the complexity of software. I understand that Intel would like to safeguard their investment in and commitment to the thread-based model of parallelism, but the Berkeley group should have the courage to tell Intel that they're full of it. Intel is making a big mistake by emphasizing threads. Same goes for AMD. Multithreading is pure folly and no abstraction layer is going to solve the many problems of threaded programming. What is needed is a universal multicore architecture that can handle anything you can throw at it. It must implement fine-grained, deterministic parallelism in an MIMD configuration. In addition, the programmer should never have to think about the cores. The processor should be completely invisible. Contrary to conventional wisdom, the programming model must be implicitly parallel and explicity sequential. That's the only way to solve the programming difficulty problem. Intel and Microsoft are wasting their money if they apply pressure on the researchers at Berkeley to accomodate Intel's ridiculous thread-based approach to parallelism. Threads are not the future of parallelism. Find out why at the link below: Nightmare on Core Street:

Like Us on Facebook
EE Times on Twitter
EE Times Twitter Feed