Thursday, 30 April 2009

Apple and chips. Apple, Intel, ARM, Imagination, ... Oracle and ... Sun

I keep coming back to it. The question keeps nagging away at me. What is Apple doing building up a chip division? Why has it spent money on ex P.A. Semi engineers? Why is it still advertising for chip engineers? What on earth is it up to?

The consensus amongst Apple-watchers seems to be that this has to do with the mobile side: the iPhone, iPods and whatever sundry network appliances make it out of the gates in the years to come. The desktop and server markets are just too firmly owned by Intel. Apple, having not so long ago come in from the IBM Power PC sphere where it had been getting dangerously far behind in terms of performance, would never be foolish enough to walk out of it and risk that happening again. Plus, Apple had to make a significant and presumably costly effort to port its software from Power to Intel, and would not soon countenance another such expense.

So on the mobile side, the iPhone has an ARM cpu (much criticised by Intel as being slow and lacking horsepower) and a couple of Imagination Technologies' PowerVR graphics processors (a geometry engine and a 3D accelerator). iPods also have an ARM based cpu; I'm not sure what gpu they have, if any.

Most commentators seem to think that this is all about adding unique features to Apple products, and reducing the amount that external chip suppliers know about Apple's technology. I think maybe it's also about taking out stuff that Apple don't need: currently they are buying in complete, general-purpose, mass-market, chips. For example the iPhone cpu, a Samsung device based on ARM IP, contains a "Jazelle" subsystem meant to accelerate Java performance on mobiles: since Java performance on iPhone has zero value for Apple, that's just so much wasted silicon, wasted money and wasted power.

It's possible then that Apple want to design their own chips. Now ARM technology can be licensed as IP, not just bought in as prebuilt chips. Apple could bypass Samsung and license ARM core IP directly. Exactly the same approach is followed by Imagination, so Apple could license the necessary PowerVR IP from them, and go on to design a single chip that incorporated not only ARM and Imagination cores but also their own, unique goodies. Even if all they end up doing is integrating into one or two chips stuff that they could have got elsewhere in several, the resulting savings in manufacturing costs (and more importantly, power consumption) would be well worth having.

This has to be far and away the most likely scenario, so it looks good for ARM and Imagination, less good for companies currently supplying the actual chips. Though Apple would undoubtedly farm out its chip making to a foundry, so perhaps one of its chip suppliers would benefit there.

Of course, a speculative post like this one wouldn't be complete without an absolutely outrageous coda. So, I'll go there. Pretty soon, Steve's very good friend Larry is going to have a RISC chip making business that I suspect he's not likely to really want. Apparently, Sparc chips can be implemented with an extremely low logic gate count — the same order of magnitude as ARM designs. A couple of weeks ago I wondered why SUN had never scaled Sparc down as well as up (the name stands for "Scalable Processor Architecture" after all). Perhaps Apple might be prepared to do that? Maybe not for the iPhone/iPod, but possibly for the tablet/netbook markets, the sort of space that Intel is aiming for with Atom, or even for the home server market?

Sunday, 26 April 2009

Most unexpected thing so far today

Following a link from Silicon Alley Insider, I end up at Microsoft's Wonderwall site. I'd seen it before and thought it was rather slick. "Damn fine advert for Silverlight," I muttered, and right-clicked on the scrolling video-wall to examine its properties.

Oops! — "About Adobe Flash Player 10...". How very embarrassing for Microsoft!

Monday, 6 April 2009

Sun / IBM deal to collapse?

The hot news of the hour seems to be that IBM's bid for SUN is off — they couldn't agree on price. This possibility has been discussed at length in the tech press in the last few days, and the consensus is that this is going to be a disaster for SUN: Schwartz has supposedly hawked it around the whole world looking for a buyer, and (supposedly again) no one expressed interest apart from IBM.

The fear is that, now they know that SUN's top managment has effectively admitted it's unable to return the company to profitability, both customers and employees may haemorrhage away. Some have even hinted that this was IBM's game plan all along — which is not an impossibility, since that used to be Microsoft's favourite trick.

There is perhaps, however, one last, desperate hope: Cisco. The investment boards are humming with the idea that SUN is a perfect buy for Cisco: it would give them an 'in' to the server market, which they have so very recently decided to enter. Hmm, Cisco's new boxes are all Intel I think, and while SUN has a substantial Intel business, its main line is of course Sparc. That said, Cisco is nothing if not an adventurous and technically competent company, and if they decided to run with the Sparc heritage, you might end up seeing a version of the chip ending up in their routers :)))

Sparc has long been almost a millstone round SUN's neck. I'm not quite sure why. ARM has made a major success of a non-Intel chip, with an enormous number of licencees busily fabbing away and integrating it into their custom designs, and I know that SUN has been willing to make the IP available to fabricators in what seems to me a similar way (the devil's in the details though) but so far, only Fujitsu has taken it up. Maybe they were myopic about targetting the architecture at the server market: maybe they should have focused more on scaling Sparc down as well as up. They did that a little bit, aiming it at multi-chip webservers, but perhaps they could have gone a bit further than that.

Whatever. It's now looking, with the benefit of hindsight, as though Java has been an enormously costly mistake for SUN. Not that it hasn't been a very successful platform — it has. But because they've never managed to make more than small change from it, while it's sucked up an enormous amount of management (and, in the early days, legal) time and effort that would perhaps have been better spent on SUN's core competencies.

I say that with a good deal of regret, and as a fan of Java. And of course, it didn't look like that in the early days. Java, originally developed for set top boxes, came out at a time when Microsoft ruled the desktop even more than now, when Microsoft had all the mind share, and when it had started to stop making versions of Windows for desktop and server class processors other than Intel's. The worry was that a Microsoft future would be an Intel future, and that Sparc would go the way of MIPS. Java then, with its virtual machine and processor-agnosticity, was a way of keeping Sparc in the game. But somewhere along the line it got too important, and SUN ended up chasing a mirage.

The brave investor then, might wait for SUN to fall below $4, even as far as $3, and then buy in the hope of a Cisco offer. Or even the hope of the board kicking Schwartz out the door, which would be bound to have a salutary effect on the share price, even if only temporarily.