In a conversation last week, ZDnet's senior editor, David Berlind, asked me what would constitute an exciting development for x86 - and I didn't have a thought to offer.
But that was then, and now? consider that nobody makes and sells an x86 CPU optimised for Unix -i.e. an x86/SSE instruction set processor with no backward compatibility other than what's needed to provide BIOS boot support and work in a standard PC motherboard?
The last time Intel tried to re-invent x86, with the Pentium Pro for Windows NT, the performance results were spectacular but Microsoft's late delivery on NT coupled with AMD's inclusion of 16bit compatibility in its competing line forced Intel back to backwards compatibility for the Pentium II - just as Microsoft's failure to deliver on Itanium has combined with AMD's x64 extensions to all but kill that processor now.
But is there an opportunity today?
Two thoughts come to mind. First, there's Transmeta. I have no experience with the Crusoe design but know it's actually a VLIW core with software instead of hardware handling x86 translation - and therefore wonder whether code couldn't be compiled directly for the processor? thus enabling a system in which legacy x86 and native applications could be mixed with the native stuff running much faster and at lower input power than on a "real" x86 processor.
The other thought is that Intel may have a unique opportunity to reduce its reliance on Microsoft by going after the Unix market with a dedicated x86/SEE instruction set processor with no legacy components at all.
That would require some kernel code to boot and run from the traditional 16bit BIOS, but Intel could provide that under GPL and thereby catch Microsoft between a rock and a hard place - just as Microsoft's commitment to Xenon as a possible x86 replacement is a threat to Intel.
Since Intel should be able to get such a chip, with the required code and possible extensions like on-board cryptology and networking, out the door long before Microsoft can get its PPC based home office products in the field, this could help Intel balance its dependency relationship with Microsoft.
And since both ideas would focus on the Linux/BSD/Solaris market for early adoption, they'd both be exciting x86 developments, right?