Barney's Rubble
Super PCs Deserve a Super OS
Software still has a ways to go before it can do justice to today's supercomputers.
For decades Moore's law stood and the density of transistors doubled every
two years -- just like Gordon predicted 30 years ago.
But even as we regularly doubled our transistor counts, the chips remained
based on a monolithic, single microprocessor model.
Now the hardware geniuses are adding some pretty wild twists. We all know about
multi-core processors (an area where AMD has been nicely assertive). By 2015,
Intel says it could have processors boasting hundreds of cores.
The graphics chipmakers are doing even wilder work. NVIDIA recently announced
a new line of Tesla boards (I'm not sure if these are named after the bad '80s
hair band or the inventor of the AC/DC current) that come with as many as 128
parallel processors.
My annoyance is overshadowing my excitement. Software historically has been
able to sap the strength of the fastest new chips. But let's face it. Much of
this software was bloatware doing things we never asked it to do in the first
place.
Now hardware is firmly in the lead. Even the cycle-sucking combo of Windows
and Office will have a hard time stressing an AMD quad-core, Tesla-equipped
PC.
Unfortunately, my last point is purely theoretical. Right now, Windows is designed
mostly to exploit our old style of processing. To truly exploit PC-style supercomputing,
brush up on your C programming skills. Many of these systems are built for scientists
and engineers who don't mind getting down and dirty with code.
Microsoft can solve this problem, but it involves changing the very way software
is written against Windows. First it has to get over its fear of 64-bit and
go nuts supporting state-of-the-art processors. And not just Opterons and Athlons
and Itaniums (remember that one?), but the Power6 processor, which can now boast
the world's fastest microprocessor and drives the fastest server and supercomputer
as well.
Microsoft, especially Microsoft Research, is working on these problems. These
efforts seem way more aimed at specialized applications, rather than transforming
the fundamental way Windows works with back-end hardware.
However, there are companies built to solve this. Google recently bought PeakStream
(perhaps to optimize its own server farms). While I don't see Google apps as
a threat to Redmond, if it can promote a PeakStream API to developers, our old
friend Windows could be very much under siege.
Whether Google or Microsoft take the plunge, either way we may eventually be
able to reinvent what PCs are capable of!
What kind of computer power are you jonesin' for? Let us all know by writing
[email protected].
About the Author
Doug Barney is editor in chief of Redmond magazine and the VP, editorial director of Redmond Media Group.