To simplify software's access to multicore parallelism, AMD proposes new x86 spec. Will Intel accept?

For more than a decade, where ever Intel went with the x86 instruction set, AMD followed. And, ever so briefly, when AMD broke ranks with its 64-bit extensions to that instruction set, Intel followed it (there's a fairly broad cross-licensing agreement between the two when it comes to the x86 architecture). The benefits of an x86 standard mostly accrue to the software companies that make operating systems and the development tools for writing software to run on those operating systems. For example, Microsoft.

If the team within Microsoft that's responsible for its Visual Studio development tools only has to worry about one way of making 64 bit technologies available to developers, not only does that make life easier for Microsoft, it also doesn't put Microsoft in the awkward position of having to prioritize one approach over another when it comes to releasing new versions or updates of Visual Studio.

Should AMD and Intel choose to extend the x86 instructions set differently for the same thing, the burden falls upon the shoulders of the OS and development tool makers to reconcile those differences in their tools in such a way that developers only see it as one way. Until recently, it seemed as though Intel and AMD were content to stay in lockstep when it came to extending the x86 instruction set. But then virtualization came along and, for the first time, the x86 instruction set experienced true divergence around a very core function (there might have been other dalliances, but none that I can remember around such a core function).

Perhaps realizing that such bifurcation doesn't serve anybody in the industry very well, AMD today announced that it's seeking to generate some consensus around standard approaches to x86 parallelism, particularly as multi-core technologies head north of four cores to eight and more. The problem, as AMD's vice president of software engineering Earl Stahl explained it to me, is that its relatively impossible for most software developers to harness the advanced performance features of today's chip architectures unless the chip and platform vendors get together to make those features transparently accessible to developers.

Already today, there's a dearth of software out there that's fully prepared to leverage the 64-bit technologies found in the x86 chips that come from both companies. Several years ago, all the 64-bit proponents could talk about is how, somewhere down the road, we'd see these incredibly rich user interfaces and games would be taken to a entirely new level, compliments of the new 64-bit capabilities. Today, apart from some nice glassy effects in Windows Vista (hardly a game changer in the big picture), that promise remains largely underdelivered.

If you think chips have outpaced the performance needs of software so far, things can only get worse if the majority of horsepower available through multicore architectures remains untapped. To solve that problem, AMD is hoping that it and the other major players in the x86 ecosystem (Intel, Microsoft, Sun [with Java], etc.) will want to work together towards some common standard rather than apart as has been the case with x86 virtualization. As you'll hear in my podcast interview of Stahl (press the play button above, download it, or if you're subscribed to my IT Matters series of podcasts [Here's how to], it will show up on your PC and/or MP3 player automatically), AMD is putting forth the first of what could be several extensions regarding parallelism and is looking for feedback from the other major players in hopes of eventually arriving at a mutually authored standard that can be incorporated into chips from both AMD and Intel.