SMALL BIZ ADVICE
What would you ask Michael?
The Dell Community has contributed: 9794 ideas | promoted 668271 times | 76278 comments

How IdeaStorm Works:

Post Promote Discuss See
Desktops and Laptops Comments | Desktops and Laptops Ideas
10

ARM PCs

10 points posted to Desktops and Laptops by bluefoxicy 05/15/07

The ARM architecture is low-power, high-performance embedded processor architecture. Used to be you could get an ARM-based desktop running Debian from some place (I believe Intel manufactured the processor and board), but not any more that I can find.

There's been talk about Ubuntu on Dells. A low-power ARM-based laptop would be an interesting offer; Debian has an ARM port, it would probably take a release cycle to get Ubuntu to produce a beta ARM port but this may lead to lower power consuming laptops.

ARM uses a completely RISC architecture and can execute instructions more quickly for better performance at low clock speeds; because of this, ARM can consume less power and still supply good performance.

Iyonix seems to supply machines at http://www.iyonix.com/ but no laptops, and choices are limited. Reviews say the measely 600MHz single-core ARM comes up fairly quickly (http://www.iconbar.com/forums/viewthread.php?newsid=919) I'm curious what a dual-core Xscale or dual 600MHz chips would do myself..

The downside to ARM is it won't run Windows. At all. Windows programs won't run under Wine, x86 Linux apps won't run unless you use Qemu. Still, there's a lot of native Linux apps; for those of us who will use straight open source or wrapper anything x86 with Qemu, a Linux-based ARM could prove quite attractive.

jmxz
05/15/07		 The downside to ARM is it won't run Windows.

Of course it can - that's what the Dell AXIM was.

I think the problem is that Windows is too bloated to do it well.
bluefoxicy
05/15/07		 Ah, I guess it could. But I'm more talking about installing WinXP Home on the laptop, loading Quake 3, and playing for a little while. To do that you'd need an ARM/VT hardware virtualization architecture with a MicroKernel hypervisor with JIT built-in that runs the kernel by re-instructioning it to the ARM from x86 (and other technobabbly things that won't work very well anyway).
bluefoxicy
05/16/07		 So there's a lot of votes here against, and a lot for (almost keeping this balanced but it's fallen to -20 as of this post).

Nobody seems to be commenting specifically on why though. My assumption is that the votes are split between "I have no idea what you're talking about but it sounds good" and "oh God another Mac-thing with their own CPU that doesn't run Windows." Must just be another uninteresting topic...
jmxz
05/16/07		 I think some of the against votes are from people who'd rather see other Dell platforms than ARM first. I think the IBM CELL chip got quite a few more votes up than down -- and indeed I'd be more interested in a CELL based computer than an ARM one. Yeah, I know those'd be totally different markets - but it's the same dell resources they'd probably need.
rebanyo
05/17/07		 I like the idea. It sounds interesting.
jorge
05/17/07		 We don't need a processor that makes Windows even slower do we? Ok you say install Linux, still slow!
bluefoxicy
05/22/07		 CELL is very complex to program for, it's not quite general purpose computing; to pull its full power requires low-level super computer programming really. It's got 8 sub-processors and one main processor that work together in specific ways.. awesome for scientific applications, some specialized workloads (it's like a GPU, it does some non-generic tasks like 3D rendering super well), but not for generic tasks unless you put some real tweak-time into it (might as well code all assembly).

ARM is faster than x86 clock-for-clock, by a lot. As for making Windows even slower, yes it will ;) But not Linux for ARM. My main concern is power though; the RISC architecture allows much less dye space to be used, so much less power for a simpler chip. The GBA used an ARM and could run 10-20 hours straight on a pair of AAs, don't think modern day processors would do the same at the same clock (nowhere near it; even at the low clock you need a hefty vcore to push the power through the semiconductor, just a lot less than at full clockrate).

I hadn't seen the CELL one, totally forgot all the hype that thing got :) To me using the CELL as a CPU has always seemed akin to using a Geforce 6800 as a main CPU....
jmxz
05/22/07		 bluefoxicy: "akin to using a Geforce 6800 as a main CPU"

Note that doing so can give spectacular results for some applications. One interesting one is the GPUSORT project which runs a sorting algorithm in the GPU. http://gamma.cs.unc.edu/GPUSORT/results.html
To your 6800 example, it's not too impressive:
"NVIDIA GeForce 6800 GPUs indicate better performance in comparison to the qsort and stlsort routines on a 3.4 GHz Pentium IV PC."
but other GPUs fare even better.

Regarding programming complexity --- note that Intel's MMX is hard enough to program that very people do that as well. In each case (MMX & Cell & GPUs) a few library authors will need to know the details; and the rest of the programmers will just use the libraries.

For the CELL, I'd expect it'd be useful most anywhere MMX is useful -- i.e. doing batches of repetitive math.
bluefoxicy
06/10/07		 jmxz: You're right, of course. The GPUs will do specialized tasks extremely well; and a few low-level specialized libraries will take advantage of them FOR SPECIALIZED TASKS. I'm not sure how generally useful that would be really; micro-benchmarks are pretty useless in general, remember. Though, to be fair, the only tasks that really need it are software 3D rendering; image processing; audio and video encoding; compression; and compiling Gentoo. Everything else is pretty much real-time event-driven stuff that uses the CPU for a few hundred thousand cycles and then goes away (like clicking a button, or opening a TCP connection and then waiting for a response from the server).
jdh30
Mar 9		 I agree completely. ARM have a great new range of higher performance low-power CPUs and they would be ideal for something competing with the Asus Eee PC. Battery life is vastly better, as you say.

Actually, that Iyonix Linux PC looks very nice. It even has an nVidia GPU. A portable version would be a Godsend...
bluefoxicy
Jul 7		 jdh you're right, with the market for Ubuntu Dell laptops AND UMPCs running Linux (like Asus EeePC), Dell could possibly do an Xscale-based UMPC running Ubuntu, *if* Ubuntu will officially support ARM (which would probably be a favorable proposition for them, since they likely would enjoy the environmental aspect of a 2x600MHz XSCale SMP system at 1 watt; of course that compares to 900MHz@5.5W in the EeePC, which btw is nice still.

Intel's new XScales at 1.2GHz have 11W TDP and the dual core version has exactly the same; the single core sets you back about $58, with $150 for the dual. Personally, I'd think 2 x 600MHz@0.5W would work better for this than 1x1.2GHz, and if I could find a price on that I might be inclined to say 3x600MHz (triangle, distance 1 between any CPU to any other) for 1.8GHz@1.5W 3 core. Two cores + Linux gives a MASSIVE responsiveness improvement; 3 cores just means when you rip a CD you can encode 3 tracks at once (faster).
 
Atom feed track comments for this idea
Please log in to post a comment
recently promoted by: