Intel Unveils New StrongARMs

mirko writes "Supported by the Epoc, Windows CE, RiscOS and VxWorks, the StrongARM RISC processor, which features power, low-consumption and high-frequency, could bring lots to the wireless market. This article and this other article describe Intel's new XScale micro-architecture that will be used in the forthcoming 1GHz StrongARMs."

Re:Just when you thought...

Errrmmm...
Done. [netbsd.org].
BTW it was damn fast with a 202MHz SA on a RiscPC, even despite its lack of FPU.

--

Intel's undue credit..

The second article says "built using Intel's StrongARM technology". Surely they mean Advanced RISC Machines Ltd [arm.com]'s StrongARM technology [arm.com], which Intel did nothing but pay some money for.

ARM is already bringing a lot

I guess the editor is probably not aware that arm processors already run a lot of deeply embedded applications. For instance, the soundchip of the dreamcast, or most cellphones. They have a 60% market share for the comms market I think. I am currently working with such a platform (lowlevel system work, and trying to fix issues in gcc as well when I can), and I must say it makes a lot of sense from a design point of view. Highest code density (especially in Thumb mode) of all 32-bit CPUs, quite efficient, and very low power consumption. No wonder it's used all over the place.

Re:RISC ?!? It's here already

Sure it sounds nice... Doens't all new tech ?

It definitely sounds nice ... I can already see the river of drool coming from people who own Risc PCs (I almost refuse to call them Acorns since Acorn sold out their workstation division but I digress).

But seriously, have anyone considered that these are RISC processors ? Do they (Intel) plan to abandon their CISC processors for the private user ? Or is this simply Intel's way of saying "we want a bigger piece of the Business pie". I certainly think the latter is true. I seriously doubt that we could get along without the CISC, it would just cause to much incompatability, or the translation matrix would make the apparent speed increase gained from the CISC->RISC insignificant. This has no bearing on "us" the private users as I see it.

I think a small history lesson is in order. The ARM architecture is not a 'new' architecture - it dates back to the mid-eighties when Acorn, having decided to skip the 16bit generation, started working on a RISC 32bit processor. In 1987, the first Acorn Archimedes was born, running Arthur OS - a fairly primitive but useable GUI and OS. This was running an ARM 2 processor at 8MHz.

Later revisions took the processor design to ARM3 with improve level 1 cache. Note these machines had no level 2 cache - as clock speed increased, this would have throttled a x86 style processor, but the ARM has fairly light memory usage as it has 13 general purpose registers, a fairly orthogonal (and small) instruction set and a load/store architecture minimising the need to go to memory for information.

Then came the ARM6 and 7 cores which took speeds up to 40MHz. At this point the ARM chips were running market leading MIPS/watt ratings - no ARM machine I have ever had has needed a heat sink - but the clock speed was starting to lag the x86 line badly. After a joint project with Digital, the StrongARM was born, screaming along at 200MHz way before the Pentiums got there, and running at less than 1W. By this time ARM Ltd had been born out of Acorn to pursue its chip dreams - but not fabrication of chips. ARM Ltd is a purely design-orientated chip creator - other partners actually build these processors. A quick trip to the ARM website will quickly show you just how widespread the ARM processor line has become - ubiquity is an almost acheived goal :-)

But just as the ARM 7's had topped out around 40MHz for a while, the 200 MHz (sometimes oc'd to 287MHz) StrongARM has remained the fastest ARM chip for a good while. During this time, DIgital got into a patent/IP dispute with Intel and ended up having to sell the StrongARM team to Intel as part of the settlement. So this is the first news of a faster ARM processor for several years - I got my 200MHz RiscPC workstation a few years ago and it blew my socks off with it's slick performance. RiscOS which is the oft preferred OS for this processor when it is used in a workstation (rather than a PDA, router, or other electronic utility) is pretty quick, and a 5x boost will be fairly insane :-) And naturally there is a port of Linux for the ARM processor (but here on Slashdot we expect nothing less).

I just regret that my RiscPC is back in the UK and I'm here in Canada with an x86. :-( Still it would almost certainly require a motherboard upgrade ...

Cheers,

Toby Haynes

More details

There is quite a lot of information on Intels website [intel.com].
But first I should mention that there is a port of Linux: ArmLinux [linux.org]
The BSD one seems to be delayed [freebsd.org].
Now, to the technical stuff.
According to Intels site, they have added power management features to the chip that allow the clock speed to be adjusted from software. Yes, this is similar to Cruesoe, but it seems like they have taken the concept even further, allowing one to go from 0 (standby) to 1000MHz. Not bad.
They have also added a few DSP functions for multimedia applications. Further details:

• 7/8 stage superpiplined core
• 32KB Instruction, 32KB Data Caches.
• 2KB Mini-Data Cache to avoid data cache thrashing
• Performance Monitoring Unit (Like on PII and up)

Please fix Slashcode

Hey CmdrTaco, there's a bug in the Slashcode.

It seems that every few days or so yet another Intel story will crop up, but instead of being a real story about chips shipping or computers being made with them or actual R&D they're doing, it'll be an Announcement or Press Release about Something Really Cool (TM) that Intel will be making, shipping, and selling Real Soon Now (TM).

Please have a look and see if you can fix this. Thanks.

Re:RISC ?!? It's here already

You're mostly correct Toby, except for:
>During this time, DIgital got into a patent/IP
>dispute with Intel and ended up having to sell
>the StrongARM team to Intel as part of the
>settlement.

DEC did not HAVE to sell the fab that made Alpha's, StrongARM's, and PCI bridges. Although we may never know the details, the agreement between DEC and Intel was that in exchange for DEC not sueing the pants off Intel for patent infringment, Intel had to take the fab off of DEC's hands. Intel got the deal of the century. They got a first class fab, the PCI bridge chip business, AND StrongARM. Too bad for them tho that the original StrongARM team up and quit enmass.

Many within DEC thought that the CEO at the time was just prepping DEC for sale to Compaq. The rumours were that DEC had both Intel AND Microsoft dead to rights on technology that DEC developed (Alpha and VMS) and that Intel and Microsoft aquired without license.

As I said, we'll probably never know the details. And that's a shame.

Estimation of pentium equivalent

I've done quite a bit of development on StrongARMs. I find that for typical workstation work I would multiply the StrongARM clock speed by 0.60 or so to get PentiumIII equivalents. That is, a 200MHz StrongARM performs about like a mythical 120MHz PentiumIII.

That 0.60 is variable by task. Some tasks its closer to 1.0, some its closer to 0.10 (we just don't do those :-)

So, lets look at the Intel XScale Benchmarks [intel.com] or rather hallucinatory benchmarks since they don't have the silicon yet...

It looks like a 1GHz Xscale is about three times as fast as a 233MHz StrongARM. Thats three times as fast as a Pentium III 130MHz, or perhaps in the ballpark of a PentiumIII 400MHz.

So stop having GHz envy and instead marvel at the really neat parts of the architecture.

• Its a PIII-400 eqivalent at 1.6Watts. Heck, the fan only on your PIII-400 is probably 1.6Watts. Even a Beowulf cluster of these won't make enough heat to keep Natalie Portman's grits hot.
  
• It probably costs 10-20% what a PIII costs. There will come a time (has come?) when you just don't need the fastest CPU to run office productivity applications or surf the web.
  

(Incidentally, don't b*tch at me about my 0.60 estimate. email me and I'll give you an account on a benchmark machine and you can't run your own. If you can figure out how to email me then you aren't worthy.)

Multiprocessor boards?

One poster wrote that the StrongARM is only about .60 the speed of a PentiumIII at the same clock. That number varied and would even go down to .10.

But at 1.6W and only costing around $20, how many of these could you put together to summarily spank the Pentium performance-wise and still run cooler and leave more in my pocket? Are there any workstations that already have multiple StrongARMs?