Conference season is in full swing; this week the Intel Developer Forum is generating a number of headlines. Intel CEO Paul Ottelini opened the show Tuesday by previewing the company's next generation architecture—its biggest product shift since the Pentium 4. But the new hardware strategy depends on software developers for success.

Intel's already looking past mere dual-core chips and planning for quad-core and even higher processor densities. Of course that won't be possible unless power requirements and heat output are slashed, so efficiency is the name of the game going forward. New chips should be judged not by clock speed, but by performance per watt, Ottelini suggested. Intel's product roadmap calls for new dual-core mobile (code named "Yonah"), desktop ("Presler") and server ("Dempsey") chips to come out early next year; these will be followed by low-voltage processors for each platform by the end of 2006.

But the thing is, Intel had a lot easier time marketing its chips to consumers back in the days when clock speed was everything. Now the company has a real vulnerability: unless programmers write applications that take advantage of multi-core chips, customers won't see the benefits of upgrading. Intel's been beating that "multithreading" drum for a while now, but you can expect the rhythm to grow more frantic. The company will do anything in its power to get developers to thread their apps, and that'll mean more tools and better support. Intel's VTune and C++ Compiler have been upgraded, there's a new Threading Tools kit with a thread checker and thread profilers, and the company is upgrading its developer education programs. You can check it all out on Intel's Software Network Home.

One of the interesting tidbits on that site is an interview with Dr. Richard Wirt, Intel Senior Fellow, corporate vice president and general manager of the Software and Solutions Group. He says there are 10,000 software developers at Intel, and that the company has a bigger compiler team than Microsoft. He notes: "Developers should think of this as a shift to multi-core architectures. Following Moore's Law, every 18 months to two years we'll be adding more cores to a single die. Code should be able to dynamically utilize any number of cores and programmers should employ techniques to generate many threads, not just two. Code will live for eight or 10 years and you don't want to have to go back and rewrite code as chipmakers begin placing four, eight, 16 or more cores on single die...Using the OpenMP specification—a set of compiler directives, library routines, and environment variables to specify shared memory parallelism in Fortran and C/C++ programs—can help in writing scalable, threaded code."

He also points out that there are cases where even non-threaded applications will see a performance improvement on multi-core platforms, but the way I see it, those incidental gains won't be enough to sustain Intel's sales. If the next few years go by and developers don't really commit to threading, the company will be in trouble. And while that might be a worrisome fact for Intel's management, it's good news for programmers, who only stand to benefit from the hardware maker's software dilemma.

Posted by Shannon Cochran at 01:43 PM | Permalink