Intel in performance-per-watt turnaround - RISC OS News, Software and Information

Intel in performance-per-watt turnaround

Published: 24th Aug 2005, 05:40:15GMT Source: drobe.co.uk
By Chris Williams

Page 1 of 1

PDA XScale family bursts 1GHz barrier

An Intel staffer puts the finishing touches to the CEO's speech

Intel has revealed plans for gigahertz busting XScales and increased performance versus power consumption at this week's Intel Developer Forum. A prototype XScale processor family, codenamed 'Monahans' and set to supersede the current PXA27x group, was demonstrated running at 1.25GHz and rendering high definition video to a PDA sized display. Sean Maloney, VP and General Manager of Intel's Mobility Group, said the chip will be launched in the final quarter of 2005, whilst admitting that the processor may not be introduced at gigahertz speeds, although later parts could certainly go this high.

None the less, it's a breakthrough for the XScale range. Although the PDA family of XScale processors are built for mobile applications and therefore do not include interfaces such as PCI that the Iyonix's Intel IOP80321 features, they do pack USB, wireless networking, Flash memory storage access and related, useful functionality. This brings the expected Monahan specification closer to mobile processors that feature ARM cores, as seen in the A9 range. Aimed at smartphones and hand held computers, the new XScale processor family is "expected to provide five times more performance within the next few years, while consuming less energy than previous Intel-based platforms," according to Intel.

This all forms part of Intel's drive to turn around its image as a manufacturer of space heating processors and concentrate on creating more efficient chips. As the computer world hurtles towards becoming increasingly mobile, it's now painfully obvious that current guzzling, battery eating processors are not going to do Intel's bottom line any favours. As any physics professor will tell you, chip makers are starting to hit the limits of silicon, where continually shrinking gate sizes is no longer a long term option. While boffins work on improving dielectrics and other material properties, there are other tricks up the sleeves of Intel's engineers: for example, one technique to save power is to only enable selected areas of a processor's circuits as and when required by the code it is executing. Another trick is to design the processor so that it consists of a number of lower power cores that can execute code rapidly in parallel.

The overall effect is that x86 processors are, at least according to the hype glowing from this week's IDF, now about to drop in terms of power consumption requirements whilst boosting performance. The new 2GHz x86 mobile cpu codenamed 'Merom' is expected to draw 5 Watts, for instance. It's still not a patch on the IOP80321, which draws about half a Watt, and cores such as the ARM9, which draw milliwatts. However, it's a pretty surprising U-turn for the makers of beasts including the Pentium, who now ideally want Windows Vista running on 0.5 Watt processors by 2010.

Incidentally, Intel also demonstrated a 'rugged' PC that can be used in developing countries and survive harsh environments. It all sounds like something very familiar indeed.

Links
Intel intros 1.25GHz XScale from The Register
Intel Merom is designed from the ground up and Intel showcases "rugged" PC for developing countries from INQ

This article has been linked to, or is available in the following formats:

[Printable] [Digg this] [Blog search]

CJE 24/8/05 11:55AM	I can't find mention of Vista running on the new processors on any of the links given. If they are talking about full blown Windows and not Windows CE (new name I forget) then that would be very interesting for us as it should be then possible to build a desktop machine (i.e. hi-res graphics) runing RISC OS
ninja 24/8/05 12:07PM	I assumed the talk of Vista was referring to running it on an x86 processor which consumes the same sort of power as the current IOP80321. That is, proper windows. If they were talking about running CE on a .5W processor, then a 2010 goal would be rather too easy.
bucksboy 24/8/05 12:27PM	It seems pretty clear from The Register article (and named reference articles) that the .5W processor in question will be x86, not X-Scale.
druck (+4.7) 24/8/05 1:29PM	Forget about that x86 nonsense, we aren't interested in running Windows "Vista". The worringy aspect is the new X-Scale variant is claimed to only have a 25% performance improvement from a 50% clock speed increment. So while in x86 land Intel may be seeing the light moving and away from the "clock bitch" Netburst architecture of the P4 back to the more efficent PIII like architecture used in the Pentium M and the up coming (as in in years behind AMD) dual core chips, when it comes to the X-Scale they are persuing the same old dead end strategy of stretching the pipeline and upping the headline clock speed in return for very little real performance advantage. While we'd all like a multi GHz Iyonix, the chip has to deliver a significant real performance improvement to make it worth while Castle or any other manufacturer using it. I've always been less than impressed with Intels much delayed and meagre ongoing development on the X-Scale, after they killed DEC and squandered the head start achived by the StrongARM. Now ARM easily has the superior core designs, delivering more instructions per clock and watt in the ARM10 and 11, but unfortunately highly integrated SOCs based on these cores haven't yet been demanded by the market, so machines such as the A9Home are based on the older but still impressive ARM9 cores.
em2ac 24/8/05 5:04PM	It seems to me that Intel are using the ARM to just learn from it, as opposed to developing it? By now I was expecting to see a 3 GHz chip...but alas everything is on the sub 1Gig side :@( not good. Just imagine if they used the same processes / manufactureing Techniques to speed up the ARM Chip. Obviously the are not interested.
ninja (+2.5) 24/8/05 6:20PM	Remember that the XScale is not designed as a desktop processor, regardless of what hardware manufacturers might be doing with it. It's not even a laptop processor. I don't know how much it would cost to produce a 3GHz part, but I do know that it would be a waste of money and more importantly, effort, as very few people would buy it. It would consume too much power for most existing and potential XScale customers.
BrianH (+1.5) 25/8/05 4:41AM	Another Register article appears to suggest that Intel are likely to ditch XScale development - they have to pay royalties to ARM apart from anything else. Article at [Link: www.theregister.co.uk]
druck (+1.5) 25/8/05 9:21AM	Underlines Intels never better than half hearted support of the architecture, for which their biggest contribution has been renaming StrongARM to X-Scale - classic "not invented here" Syndrome. But shoving x86 crud chewers in to nasty little Microsoft Pocket PC PDAs will be another reason to buy Palm instead.
JGZimmerle 25/8/05 10:18AM	They will probably try to do it the Microsoft way of embrace and extend. Embrace ARM (by making the fastest ARMs on the market) and extend them with their own technology, to make clients dependent on Intel. Then offer a host of migration tools for moving to their x86 platform, wich will have Intel's ARM-extensions by then, and phase out the ARM line by ending its development. Lets hope that everyone sticks to the ARM instruction set and does not use the Intel extensions.
druck 25/8/05 10:43AM	Well they've already done that implementing new co-procressor instructions in the various X-Scale cores, including MMX like instructions and a graphics accelarator in the latest PXA270. In the PDA market theses are only used by manufacturer driver writers (if even that), the extensions tend not to be used by application developers given Microsoft's stoneage development enviroment that doesn't even properly optimise for the StrongARM ARMv4, never mind X-Scale's ARMv5. But that does tend to favour the ARM9 which at only 266MHz matches a 400MHz PXA255 on almost every benchmark.