When it comes to bang per clock cycle its difficult to beat architectures like the PowerPC or the Sparc, But its obviously not what people are worried about because the x86 is charging through the pack like an out of control bulldozer --- despite it being essentially a crappy klugde of an architecture. Its probably something to do with cost effectiveness, how you get the whole hardware package for a decent cost, how you've got lots of choices you can make to get the price/performance/power point you're aiming for and the sheer volume of experience with the parts.

An x86 is low power -- provided its not actually doing anything. Same for any other processor. Its easy to be low power if all you're doing is hanging around waiting for keystrokes.

x86 is a fundamentally bad design. Even Intel has made much better chips but people were convinced they needed Windows & other apps that required x86. Yes, they did have an NT for non x86 machines but it couldn't run the apps natively.

But none of that matters with open source software. Open source will kill proprietary crap not because it's free but because it can work on much better hardware.

Even a home computer costs more to operate over it's lifetime than it does to buy. Most people don't realize that but server farmers do.

Most people commenting so far have said "no way can it compete with the Intel XYZ" missing the possibility that market dynamics can change very very rapidly.

Compact Flash is moving in on hard disks. (you don't need 1 terabyte drives for office documents, even scanned ones).

If electricity prices were to spike ARM+CF would be a good option, especially if you're the guy in the corner on the electricity-generating pedal cycle.

I'm kind of following up on Core Dude ... in that I also doubt ARM's ability to make in-roads into THE server market. Having said that, ARM does have some interesting technology and if they play their cards correctly could stand to make a nice gain.

The differentiating factor, as I see it, is that the Cell is simply not a server CPU. Niagara was also not designed to be a server chip -- it's more of a specialized communications chip (fascinating actually). Sun's Rock is a full-blown server CPU and will kick some serious Intel fanny -- performance-wise, though likely not market-wise, all while consuming insane amounts of power [the Rock runs a tad warm 8].

Virtualization would seem to be more important than low power consumption. However, low power is "green" so you at least have to pay lip service.

I'd think that any RISC based CPU is going to have an advantage over CISC design. Perhaps that over-generalizes what I'm really thinking, since you'd really want lots of cores NOT running at super high clocks. Efficiency of the instruction set architecture ... something like watt-instructions per clock. I'm sure that's already been hashed out elsewhere.

Btw, hope I didn't sound like drashek, sheesh already.

Can we stop this nonsense? IA vs non-IA battle has been fought and settled, and IA has come out as the undisputed winner. Intel itself is having trouble pushing its own Itanic line into servers. No other instruction-set architecture can challenge IA in well-established markets, let alone, the new-comer ARM. IBM tried this will Cell, Sun tried this with Niagara, and finally, both of them reverted back to good old Xeon or Opteron.

Low power? Who cares about low power, when each DIMM consumes 5 watts? Or each HDD consumes 15 watts? Have you ever heard of Amdhal's law? How in the world ARM aims to create synthesizable designs to compete with Nehalem MP?

Get real....