The Intel XScale conundrum | Drobe.co.uk archives

The Intel XScale conundrum

By Chris Williams. Published: 25th Aug 2005, 13:27:35 | Permalink | Printable

Chin up, could be worse

As if trying to reconcile some great conflict, IT pundits have been trying to get their heads around Intel's XScale and low power x86 architecture plans. On the one hand, the chip giant has announced plans to slash power consumption for its main line of processors whilst boosting performance, and on the other hand, is talking up a new gigahertz XScale family. Surely, the two cannot exist at the same time?

It's been noted that the introduction of Intel's so-called, x86-based next-generation microarchitecture (embedded processors for handhelds and similar ilk), will likely marginalise Intel's own XScale line. However, it's expected that the XScale will still continue to have a role at Intel for the next year or so, shifting towards applications that demand mobility. What appears to be the case is that Intel would eventually prefer people to use half a watt x86 processors, and in the meantime, coax clients into trying out the low power XScale line that Intel acquired and for a while, neglected. Plus, given that ARM continues to rule the mobile landscape and that the XScale is ARM compatible, it probably makes sense for Intel to ensure that they at least keep a foothold in the realm of ARM targeted development. On top of this, Intel have sneaked MMX-like instructions into the XScale, anyway.

The end result for RISC OS users is that, essentially, nothing has changed. Castle are unwilling to upgrade the processor used in the Iyonix (the 600MHz Intel IOP321), unless a new chip suddenly arrives and shows a significant increase in performance. Besides the IOP80321, which was chosen for the Iyonix because it had relatively good performance and included PCI support, embedded processors with ARM cores can also be suitable, as demonstrated by the A9home. Even though its Samsung ARM9 CPU doesn't include PCI or AGP support, its graphics and USB capabilities are arguably enough for RISC OS - an operating system that has always been able to 'do more with less'. Ultimately, by the time Intel decides to off load its XScale department for good, there's likely to plenty of ARM compatible alternatives available. It's unlikely, though, that these will ever target the desktop or workstation class of computers.

Links

Intel in performance-per-watt turnaround

Previous: Oregano, Firefox and NetSurf reviewed
Next: Survey: Iyonix use leads VirtualRPC

Discussion

Viewing threaded comments | View comments unthreaded, listed by date | Skip to the end

This does seem curious. OK, so Intel has good (self-interest == profit) reasons for trying to migrate products to x86 instruction architectures, and also for trying to retain those developers with legacy ARM code who wouldn't switch.

But what is the actual playoff between the two architectures like? We all know the ARM architecture to be small and efficient, but how much 'smaller' (more processing/Watt) is it? We can't just compare any two random -or even the most modern- chips, because intel's ARMs lag behind in process size and so on. I guess this primarily involves the gate-count, but other system requirements, like code density, probably should be taken in to account.

Personally, I'd like to see the most efficient architecture win, which at the moment I expect to be ARM. Since Intel have belatedly decided to improve efficiency by moving to multiple cores, what is the chance of the same thing happening for ARM? It would be good if this gave the ARM-architecture an unassailable position relative to x86. This might then mean progress the other way, with desktop and supercomputer class ARM chips!

Loris on 25/8/05 3:31PM
[ Reply | Permalink | Report ]

Aren't Intel taking a backward step? ARM is RISC, x86 is CISC. By putting a CISC core into these processors, surely their efficiency, as Loris said, won't be as good as the RISC cores that ARM processors have.

It seems like Intel are trying to 'improve' their processors by overloading them with features and instructions, much like they did with the x86... anyone seen this before? We all know how power hungry the Pentium 4 is - which is at the peak of the x86 'let's stuff it full of fancy instructions which will never get used'.

Smiler on 25/8/05 4:02PM
[ Reply | Permalink | Report ]

Horses for courses, methinks. I'm still enjoying using my 202.6 MHz StrongARM and Photodesk compared to my PC laptop with 10x the processor speed, 8x the memory and still it faffs about with image editing.

harmsy on 25/8/05 4:11PM
[ Reply | Permalink | Report ]

IDF is supposed to stand for the Intel Developers Conference, but its more lile the April Fools Conference if they really beleive, or rather they expect us to beleive, they can get an x86 based processor capable of running the blubbering edifice of Windows Vista with the same power consumption as a current X-Scale.

The embedded and CE (Consumer Electronics) market is dominated by the ARM ISA, and there is very little call for x86 compatibility, or indeed skill to use it as its one of the few areas where had tuned assembler is still used for kernels and device drivers. I see this story as Intel just talking up the tired old dog of the x86 ISA in the CE field as a way of taking the heat off their thrashing by AMD on the desktop and servers.

Whilst Intel have failed to lever the X-Scale in to the lucrative mobile phone market in any volume (it remains dominated by other ARM based SOCs such as TI's O-MAP), there is still enough demand from other markets for Intel to stick with the X-Scale for a good time to come. This isn't just the visible PDA/Smartphone and portable media player market which are driving up performance requirements, but also the embedded networking controllers used in switch gear and the PCI controllers used on system bridges - which is where the X-Scale IOP321 in the Iyonix comes from. As Intel have dropped their previous processors for these segements such as the i960, ambandong the X-Scale would mean also abandoning these lucrative markets.

druck on 25/8/05 4:11PM
[ Reply | Permalink | Report ]

In reply to Simler: The RISC verses CISC debated ended long ago, as no CISC processor has run the x86 ISA since the 486. All chips since then have a RISC core that operates on microinstructions that bear absolutely no relation to the x86 instructions. The nasty nasty x86 opcodes are translated by the first few stages of the pipeline, and where as this was a significant overhead on the first pentiums, the vast transistor buget of todays processors makes the decoding insignificant in terms of resources. Thats no to say a well designed RISC instruction set (register rich, three operand) such as POWER can't thrash the x86, in terms of instruction density, instructions per clock and efficent utilisation of chip resources.

druck on 25/8/05 4:19PM
[ Reply | Permalink | Report ]

I think we've all forgotten one important player in this haven't we ?

*MICROSOFT*

Intel may figure that "low power" x86 may suit MS more than ARM and may be pitching in that direction to keep MS happy.

As far as relatively low power consumption x86 devices go these already exist (Geode for one and Crusoe for another). Intel simply moving into that space won't necessarily guarantee them a *win* they *will* have competition - also (as David Ruck said) Intel has still plenty of markets to sell xScale (so is not going to drop that chip) so much of the posturing goijng on may just be them trying to please MS or just trying to leverage some of their IP in areas they previously ignored.

x86 entering the ARM niche may well force ARM to "turn up the performance" - which from our view point may not be a bad thing.

AMS on 25/8/05 6:58PM
[ Reply | Permalink | Report ]

druck: I stand corrected - thank you

Smiler on 25/8/05 7:06PM
[ Reply | Permalink | Report ]

AMS is right to highlight Microsoft. I recall an announcement some time ago that the mobile Windows-in-ROM (whatever it's called this afternoon) would be developed only for ARM and not for x86.

Note also that the desktop market is moribund this year, with only laptops shipping in quantity worldwide. Seems to be a trend.

All this leaves Intel with two specific market openings away from their traditional heaters:

1. Sub-notebooks and handhelds running ROM Windows on ARM. Sub-notebooks are probably the area where XScale can make the most impact.

2. Laptops running Windows XP on x86. Video decoding or gaming without roasting the lap underneath or running the batteries flat while Frodo is still stuck in the dead marshes is the name of the game here.

It all makes more sense than Intel's strategies have for a very long time, IMHO. I notice they are even incorporating AMD's 64-bit instruction set extensions in their new dual-core kettle brewers. Watch their share price next year.

Cheers, Guy

steelpillow on 25/8/05 10:00PM
[ Reply | Permalink | Report ]

steelpillow: That's a very narrow view of the markets for ARM processors - from Intel or anyone else. Most ARM processors end up in things which have nothing to do with MS, or even when you might not expect ARM at all. Such markets are dominated by Linux and other embedded OSes. You can go on about MS all you want, but unless you're talking CE, it's not really relevant.

mrchocky on 26/8/05 9:59AM
[ Reply | Permalink | Report ]

What MrChocky says.

On the speed of ARMs: the Iyonix looks like an even less compelling offering at its price at 600MHz in 2005 than it did in 2003. This is reinforced even further when Intel now offers a 1.2GHz XScale chip. OK, it's not 2x as fast, it's only around 25% faster, but even so, it reduces the apparent value proposition of the 600MHz part in Castle's machine, no?

lproven on 26/8/05 12:04PM
[ Reply | Permalink | Report ]

lproven:

Let's face it, Castle must have a new machine up their sleve.....

em2ac on 26/8/05 12:58PM
[ Reply | Permalink | Report ]

Depends on the market. I suspect most people who would buy the Iyonix understand the platform enough to know what they are paying for.

If Castle is aiming at Joe Public, then the price point to system spec ratio will never attract them as there expectations have been influenced by x86 PC specs. Indeed it would look as though they are getting less but paying more.

*

The MS/Intel relationships is a good indicator as Intel and Microsofts core business will be based around XP and x86 processors, with increasing focus on mobile devices slowly changing that.

However in the short term I suspect they will need to maintain a viable, backwards compatible chip architecture. Moving to ARM powered technology for desktop machines would be an expensive proposition, not only for the chip maker and MS, but for other companies offering preipherals. Not to mention probelms with legacy support.

I think to be honest that Castle have it right, they have a proven system spec with a good customer base. Future chip upgrades will depend heavily on the production of a chip that 'fits the bill', intel at present don't seem to be moving in this direction with ARM cores.

JDC on 26/8/05 12:58PM
[ Reply | Permalink | Report ]

lproven: I am not sure if an IYONIX for 800 UKP now is a significantly less compelling offer than an IYONIX for 1300 UKP in 2003.

In 2002, Samsung announced that it has a 1,2 GHz ARM10 variant "nearly ready". Did this make the IYONIX a less compelling offer? No, because it turned out that Samsung just released a lot of hot air instead of an ARM that could actually be bought.

The Intel announcement is very similar: they talk about a 1,2 GHz XScale variant (which might - independant of the clock speed - not be too suitable for a desktop computer like the IYONIX anyway - after all, it won't have a lot of RAM bandwidth, neither will it nicely integrate something essential like PCI), but also say that the first offering will be sub-1 GHz.

Look at the A9. It features a 400 MHz Samsung ARM9 variant. Why doesn't it use the often-announced 533 MHz variant Samsung were regularly promoting? Because it is not available.

At the moment (and probably for the next 6 months at least), the only suitable processor for an IYONIX-type computer is the IOP333. However, from a marketing point of view, the "jump" from 600 MHz to 800 MHz is just not enough, even though from a technical point of view it is probable that it would offer a significantly better performance.

Unfortunately, the performance gap between available ARM variants and x86 processors seems to widen all the time. Which is bad news for the future of RISC OS on the desktop and dedicated hardware development.

Even more unfortunate is that the future of high performance ARM computing might be a multicore architecture, which doesn't exactly suit the way RISC OS is structured at the moment.

hubersn on 26/8/05 1:53PM
[ Reply | Permalink | Report ]

I think you might find that work on improving the speed of ARM cores will accelerate in the very near future. The functionality and convenience of PDA's is starting to to be attractive to laptop users. This will stimulate the demand for ever faster ARM chips (amongst others). In the PDA world ARM dominates even amongst CE powered devices.

mripley on 26/8/05 3:18PM
[ Reply | Permalink | Report ]

Agreed, however PDA's and laptops don't have PCI bus', something of an essential if ARM CPU's are to be used for desktop machiness.

My main concern is the production of future ARM chips suitable for desktop use, as the current trend seems to be moving away from this. As has already been mentioned this could cause issues with RISCOS support too.

JDC on 26/8/05 3:37PM
[ Reply | Permalink | Report ]

In reply to JDC:

Remember that the X-Scale is not a single chip, its an architecture encompasing a range of chips aimed at very different segements, the same can be said for chips featuring ARM cores in general. No current ARM chips are remotely aimed at desktop use, so they can't be said to be moving away from this. The main source of chips which are also suitable for desktop systems are those aimed at the network routers and PCI controller, which require low high data throughput. These are continuing to be developed and performance increased to keep up with demands for faster network fabric and I/O marshalling to take the load off main system processors.

druck on 26/8/05 4:16PM
[ Reply | Permalink | Report ]

Careful with the sweeping statements, you might hurt someone. PDAs (no apostrope) sometimes _do_ have PCI busses, so PCMCIA devices can be connected. But even if they didn't, big deal, the PCI controller can elsewhere.

mrchocky on 26/8/05 4:17PM
[ Reply | Permalink | Report ]

Most laptops do have a PCI bus, some even PCI Express. The video, sound, wifi and even IDE chipsets may all be connected via PCI (wifi often using the mini-pci slot), making it a very important item to have.

ksattic on 26/8/05 4:24PM
[ Reply | Permalink | Report ]

I stand corrected about the PCI bus.

In reply to druck:

I basically said that in my post, but obviously not as succinctly.

The point I suppose has been reinforced.

In reply to mrchcocky:

Fair point about the PDA's (oops that apostrophe sneaked in from somewhere!), however the PCI controller will be much more efficient if its built into the chip architecture.

I wasn't making a sweeping generalisation, the first word was 'My' It was 'my concern'. I did not say that it was going to happen, only that I was concerned it would.

JDC on 26/8/05 6:12PM
[ Reply | Permalink | Report ]

mrchocky:

Going back a post, yes most ARM processors end up elsewhere. But those markets are for the most part not well-matched to Intel's proprietary X-Scale add-ons.

My point was about the areas that /are/ well-matched to Intel's proprietary technologies and other strengths.

The OS you call CE is the same thing that I called Windows-in-ROM. Hope that clarifies the relevance (or otherwise ;/ ) of my comments.

steelpillow on 26/8/05 6:53PM
[ Reply | Permalink | Report ]

Please login before posting a comment. Use the form on the right to do so or create a free account.

Search the archives

Today's featured article

Select preview
Every feature in Select 1 covered, what's in, what's out, what's it look like and what do we think of it?

9 comments, latest by diomus on 15/3/03 1:10AM. Published: 31 Oct 2001

Random article

Geeks celebrate RISC OS time epoch
Clock-watching nerds rejoiced this week as all modern ROS machines reached a mildly baffling milestone. The OS keeps track of the current time by storing it as the number of one hundredths of a second elapsed since midnight January 1, 1900 - and this figure ticked over to 0x5000000000 in the early hours of Tuesday, provoking excitement among some.

8 comments, latest by druck on 24/11/08 9:02AM. Published: 20 Nov 2008

Useful links

News and media:
Iconbar • MyRISCOS • ArcSite • RISCOScode • ANS • C.S.A.Announce • Archive • Qercus • RiscWorld • Drag'n'Drop • GAG-News

Top developers:
RISCOS Ltd • RISC OS Open • MW Software • R-Comp • Advantage Six • VirtualAcorn

Dealers:
CJE Micros • APDL • Castle • a4 • X-Ample • Liquid Silicon • Webmonster

Usergroups:
WROCC • RONE • NKACC • IRUG • SASAUG • ROUGOL • RONWUG • MUG • WAUG • GAG • RISCOS.be

Useful:
RISCOS.org.uk • RISCOS.org • RISCOS.info • Filebase • Chris Why's Acorn/RISC OS collection • NetSurf

Non-RISC OS:
The Register • The Inquirer • Apple Insider • BBC News • Sky News • Google News • xkcd • diodesign

© 1999-2009 The Drobe Team. Some rights reserved, click here for more information
Powered by MiniDrobeCMS, based on J4U | Statistics
"Thanks for the unkind comments. Why are the young such ****s? "
Page generated in 0.1994 seconds.