ARM-Based Servers Expected by 2011

Well.. software is there. If they'll be cheap enough I might even buy one. This would appeal to people making home servers from Linksys access points and set-top boxes.. If they make hardware compatible and interchangeable (like with PC) they might succeed

Man, Intel just can't catch a break with Atom can they? Not that I'm an Atom fan anyway.

MS has released at least 10 editions of Windows CE and Mobile for ARM :-)

Linux has a huge share when it comes to servers. Good thing numerous Linux distros have been ported to Arm already.  

Me neither. I like VIA though.

Some really honestly cool news, make this your next ipad level investigation please Toms.

But I'm still curious about VLIW and EPIC based CPU architectures that are supposed to deal with the shortcomings of the RISC architecture which is what the ARM CPUs are based on.

Despite the lack of Win32 support, the lack of 64-bitness could also limit the use of ARM in server environment. At least on the disk and database servers, all the available memory can usually speed up the operations.

I don't think this is gonna happen. It's just hype.

If lower power is the new thing people will just load a server up with Atoms.

Atom performance per watt and clock sucks hard! That's the deal in fact, ARM performance is far greater in both fields. And in a server farm, this mean a lot. You want to have a CPU that produce less eat, consume less power and that is more efficient. ARM server will happen sooner than later. It simply fit too well the current server paradigm : virtualization on dynamic cluster or simply the cloud!

ARM makes all the news recently, but there's another contender: MIPS64

Anyway, ARM should invest in something more PC like. What's wrong with having a heat sink, ARM? And "normal" RAM sockets? :-)

Power consumption is indeed an important issue when it comes to big server halls and the heat issues that are involved. It is hardly news that the x86 architecture with its CISC instruction set is a very ineffective one compared to the RISC architecture. It is also a well known fact that CMOS based semiconductors consumes considerably less power than HMOS, especially when idle. The ARM CPUs have made their way into portable devices which have dramatically increased the volumes which led to lower price per unit. This is really an interesting story. Developing a CPU ought to be very expensive and it probably was a big deal when the Acorns were released. But now almost 20 years later these R&D are sunk costs that has been covered for and it takes little effort to produce it for other applications. It would be interesting to know why ARM has been so successful in portable devices and not MIPS or DEC Alpha, or PowerPC. Maybe it is proprietarism that has prevented the success of the other CPUs whereas an almost unknown CPU maker, eager to hit new markets made the success of the ARM architecture.

I strongly believe that 64-bit architecture is much preferred in servers whereas there (currently) is little desire for it in handheld units.

MIPS CPUs could also be a viable option as they also have found their way in larger scale applications such as Playstation and PSP. But it would be interesting to see how efficient they are in terms of performance per watt. The old MIPS R10000+ CPUs based on non efficient litographies may not be a good reference as "finer" litographies almost certainly yields more efficient CPUs.

The next inevitable step in the CPU architecture is the VLIW and EPIC architectures where the CPU is more optimized for parallel computing and features such as cache-prediction is done on the compiler/software side leaving wafer space for other more meaningful applications. Since these CPUs are simpler in their construction, the instructions sets are clean and homogeneous, hence they could more easily be optimized and tweaked for higher efficiency and performance in general.

A big burden that the x86 architecture is bogged down with is the backwards compatibility which makes it a very inefficient CPU. ARM did introduce a 64bit based CPU in the early 2000 called ARM64 or Jaguar, I don't know where this went. I hope that VLIW and EPIC will gain a bigger acceptance some day but the CPU developers have to be smart; Trying to make a VLIW CPU compatible with x86 is not a big hit as this really costs performance and then charge a lot for the CPU to cover for the R&D is not a really a great idea which I believe Intel have experienced with their launches of the Itanium CPUs.

It seems that ARM has been so successful in that area because they have very little competition in the ultra-low-power processor market. ARM is the only architecture that doesn't have (nor ever really had) processors that competed at the high-end. Early on, they decided to focus exclusively on the (less interesting imo) low-power market, and so it's little surprise they excel in that area.

However, just because they have the best perf/watt at the low-end doesn't mean that will translate right on over to the mid/high-end. Due to the non-linear relationship between power and performance, lower-power, lower-performing processors almost always have better perf/watt than their better performing peers, regardless of architecture. Likewise, perf/watt only tells part of the story; absolute performance still matters! Even in this "age of parallel computation", serial performance is still important in the vast majority of tasks. So even if you were able to get more aggregate performance from a ton of ARM cores than from a lesser number of x86 or POWER cores in the same power envelope, that doesn't necessarily make it more desirable. To use an extreme example, would you rather have 4096 (hypothetical super-ultra-low-power) 486s or 4 Core i7 cores at your disposal?

To rephrase the original statement: it would be interesting to know why PowerPC has been so successful in the high-performance embedded market. Obviously that's what IBM (and to a certain extent Freescale) decided to focus on. ARMs highest performance cores of today still can't compare to the PowerPC-based processors used in the Xbox 360 and PS3 of half-decade ago. It's all a matter of focus. I actually consider the game console embedded market to be the most interesting of all, because performance, power, and price all have to be balanced. It will be very telling to see which architectures wind up in the next generation of consoles.

What ever happens to Trasmeta's tech anyway.

Nvida's Tegra may be a nice toy to have in the near future... probably in a phone

Sounds good...But ARM needs to work a lot on the compatibility issues when it comes to Linux or other open source distros