Thursday, March 8, 2007

Back to The Future With Powerful FPUs, Multicore CPUs, Multithreaded Programming and More Programmer Control

The next 5-10 years maybe the year that system programmers of the last 5 years have been waiting for. After about more than two decades of clockspeed scaling every single processor release by vendors such as Intel, AMD, and others, it finally hit the wall or at least very close to it. The trend today is more core per-CPU package (a.k.a multicore) rather than more MHz per chip package. This means we move toward more and more multithreaded realm because it's about the only way to extract performance out of the CPU. This is not a trivial task. It requires a lot of work and system level understanding. From compilers to programmers, to CPU designers all of them have to have an understanding of the overall system or better yet, forecast the impact of every single decision they make during a component development, be it software or hardware. Because it's not a simple "just put more MHz" to execute the code at faster rate than it used to. Two articles from Arstechnica point out about this:

Looking into it at a larger point of view, we can view the processor in the future as supercomputer of the past lumped into a single CPU package. Therefore, the supercomputer researcher of the "past" is the key person in the future of CPU design and Software development because these guys have been tuning multiprocessor system performance for years.

.. to be continued ..
Post a Comment

No comments: