With great respect, you probably won’t be deciding the outcome of the CPU race. Not if your normal shopping list has called for backward-compatible, x86 systems that are 25% faster than the previous year’s models at about the same cost.
No offence, but you’re not the one who hires technologists to analyse CPU trends and plan four horizons ahead.
That task goes to AMD, IBM, Intel, Sun and other chipmakers.
When AMD cast its mind (named Fred Weber) further, it saw its clean room x86 trimmed back to its core and built into a shockingly well-designed machine. When IBM looked ahead, it saw ever more powerful Power and PowerPCs. When Sun looked ahead, well, who knows what Sun saw?
When Intel set its gurus to work, they saw Itanium. The world laughed at this processor that Intel could not describe in understandable language. The only things clear to all were that that Itanium was incompatible with everything else and it didn’t have as many gigahertz as Xeon.
Itanium and its progeny, known as IA-64 on Intel’s family tree, embody the right idea. IA-64 makes what Intel is doing now with x64 seem like a waste of time and money. How can one say this about the king of market share? Shouldn’t Intel be in there fighting to win its honour back from AMD?
Honour, schmonour. Intel’s got the volume business and it’ll keep it until x64 is relegated to digital watches. Chalk up the win, we say, quit brooding, and start planning for next season.
The time is right for IA-64 because, except for the AMD hitch, the future played out precisely as Intel planned. We are coming to a place where instruction sets, pipelines, RISC and CISC don’t matter. Intel rolled out the original Itanium as “the processor for the next 20 years”.
Almost. The compiler is the processor for the next 20 years, and IA-64 is a compiler’s dream date. IA-64 is exactly what we need at the mid-range and possibly, if Intel builds out the surrounding system architecture properly, the high-end as well. The future belongs to processors that are essentially big clusters of logic gates that switch very, very rapidly.
It doesn’t matter that a CPU like this is almost impossible to code to because, as Intel saw so long ago, almost nobody writes in machine language any more.
The road ahead isn’t about 8GHz Xeons or 32-core Opterons. It won’t be about hardware at all. It will be about the $100,000 commercial development suites that will perform automated, distributed build, run, observe and optimise cycles until native code flows through every possible combination of processor types. And, for another $50,000, tools will instrument a commercial app to optimise itself based on changing deployment environments.
In this scenario, it might take Microsoft three months just to build a major release of Windows, harnessing the off-hours cycles of every machine on the Redmond campus, but the result would be an OS that could chop a big shop’s system requirements by a third or more.
I’m bullish on IA-64 because a dream world of compilers that take their sweet time to build and optimise but that produce mind-blowing code will surface there first. Everything learned there will transfer to other architectures, however, and we’ll end up with a naturally occurring matrix of CPU types and deployment patterns that will provide customers with meaningful choices.
Yager is chief technologist at the InfoWorld Test Centre