Jonathan, I have to agree on the apparent niche market. On the other hand, the so-called "appliance" market may turn out to be a pretty big niche. On the other other hand, there are plenty of other cheap embedded processor solutions out there. Beats me.
VLIW has been kicked around for a pretty long time now, but the long, tortured road to market of mysterious Merced aka Itanium doesn't exactly bode well for the concept. I think Merced was originally supposed to ship in '97, it's a deeply troubled project. Maybe Transmeta is doing it better, but the "code morphing" sounds somewhat like the DEC Alpha dynamic translation approach to x86 compatibility, it was supposed to work pretty well, but it didn't do the Alpha much good. It's a bit odd, the big deal with VLIW is supposed to be fancy compilers to detect the low-level parallelism, funneling code generation through x86 architecture isn't an intuitively superior approach.
On another point, though integrated North Bridge, makes sense from an embedded point of view, that just means a memory controller + PCI bus interface, right? PCI works good, and you can't beat commodity PC memory prices. Cheap leverage on existing technology.
Cheers, Dan. |
