Last week, someone that shall remain nameless (i.e. me) made a footnote at the bottom of an article that mentioned that PPC-core market is far bigger than people realize and on the order of 70M units this year alone. I wanted to correct that, and mention that "foot tastes pretty good this time of year" -- or in other words, "sales were not that high".
For the excuse: there is an embedded core called ColdFire. When visiting Somersett, they had a large sign that said something like "over 70M sold". A Motorola executive was asked, "is that the number of PowerPC cores?", to which the answer was "Yes". He probably just wasn't paying attention.
When asking Motorola marketing about their sales volume for all PowerPC cores sold either this year, or ever (to correct my error), they deftly dodged the question. I got the impression that they don't want to the competition to really know how big the market really is -- but that makes it pretty hard for their partners and allies to know either. They pointed me to some analysts.
After talking to an analyst or two, a close approximation seems to be roughly 7 Million cores per year. So I was only off by an order of magnitude or so.... does that count as both feet? But hey, that is life -- learn from your mistakes, and go on. It wasn't really the main point of my article, but heck, truth matters.
Does that 7 Million per year mean that PowerPC is doomed?
The problem is that some are implying that volumes this low mean that the PowerPC is doomed. That is not true. To understand the full complexities of the situation, I have to explain a little about the markets.
Embedded controllers are a strange market. The volumes are huge, and huge over a long period of time. Companies do things like take 5 year contracts for a per unit loss -- but they make it up in volume. I know that sounds like a joke, but it isn't.
Moores Law (really just a prediction) is that transistor technology will double the space efficiency every 18-24 months -- which drives down costs, heat, and increase performance. This has become such a constant, that companies will actually gamble on it. So some companies will take bids for controllers that they will lose money on for the first year (or two) -- but in a 5 year contract (or product) they can make it up on the hind end. They are betting that costs will go down enough that by the third year they are at least breaking even, and the 4th and 5th years are all gravy. Also volumes (and sales) usually ramp UP, nor down. (Some of this stuff happens in the processor market too -- but not as much). Still, imagine always betting on the future to dig you out of your present -- scary stuff. It's like credit card debt, buying a house, or being any normal overextended American.
There is no doubt that Motorola and IBM's future right now is based on PowerPC for the future. Already the main processor market are shifted from 68000's (for Motorola) to PowerPCs... and now the embedded market is starting to shift as well.
The embedded market moves slower than the main market. It will take years to make that shift -- Motorola is still selling 68000's in the embedded space (about 70 Million this year). Motorola will probably be producing 68K's for another decade. But there is no doubt that they want to keep those people who are shifting away from 68000's, to shift to PowerPC's. So the potential market is well over 70 Million units per year. Motorola needs that PowerPC. So I think by looking at the past, and embedded controllers, we can see that the future is clear -- PowerPC's will be around a long time. Decades more . Remember, the long term commitments of those embedded markets is huge. Motorola has to bet big with the PowerPC's or they will lose that market of 70 Million units a year to someone else.
The market realities
Some nay sayers are stating that 7 Million machines a year (with ten times that as a conservative market cap) is not enough to encourage investment in the PowerPC -- that to me sounds a bit delusional. This industry works with "Up our out". Either you get faster, or you get trampled -- it's like running with the Bulls (serious motivation to keep running).
The PowerPC is the vehicle by which IBM and Motorola will get better -- and they know it. IBM and Motorola will continue to pour money into R&D, or they can turn their markets over to someone who will. That is just reality. Some technology companies do curl-up and die -- but I don't think Motorola or IBM are in that place just yet.
The only think they are teetering on is whether they are going to completely focus on the embedded market or continue to focus on the mainstream processor market.
The mainstream market is based around making fast processors, using cutting edge technologies and processes, and pushing technology forward (using instruction set improvements, new pipelining techniques, and whatever it takes to compete in performance). It also requires more manufacturing process development, alliances, and so on, to keep up. Keep on moving, or get run over. That is your choice.
The embedded market is a tad more laid back, and a bit more into lower power -- but it is still either make fast processors or get trampled. Keep on moving, or get run over. If you really push size and cost down, and performance up, then you can survive. Most of that is applicable to making a great low-power high performing main processor as well.
I believe that IBM understands the market realities, and I am pretty sure that Motorola does. So the question is not "if" IBM and Motorola will support the PowerPC -- just which way will they go?
IBM pushes R&D harder than most companies. IBM's flops of the 80's and early 90's scared them into being hungry again, and they have done a great job of innovating. IBM will continue to innovate on their own -- mainstream PowerPC's, proprietary PowerPC's, and embedded PowerPC's. With or without Apple, IBM needs fast chips for their own business. Lots of chips from IBM, in all spaces of the market, that Apple can take or leave.
Apple and IBM haven't dealt too well together (past) because they are both big, bureaucratic companies, with large ego's (and individual personalities with larger ego's). IBM wants to serve IBM's interests first -- which are 64 bit workstation type PowerPC's. So far that has not been Apple's best interests. But Apple is pushing for Unix based Mac OS X. In another year or three, Apple's OS and focus will be shifted to the exact same kind of processors that IBM is using on its high end. It is likely that high end Macs will be able to penetrate higher-end server type markets, because of IBM (and Mac OS X). This is all by Apple riding on IBM's coat tails. And IBM is not solely focused on high end. They have some cutting edge low-power embedded controller type processors as well. (Which are great for portables, low cost machines, handheld computers and so on). There will still be many other PowerPC flavors for lower end, that Apple would be able to use as well. I suspect that Apple will do fine, even if they had to ride IBM's ego, wherever IBM took the market. And IBM is still only a fraction of the PowerPC market.
Motorola was falling behind with the 68000's (stagnating) when the PowerPC alliance started. I really believe that without the PowerPC, Motorola would have become irrelevant -- they needed the competition. Competition make Motorola hungry -- I think many inside Motorola agree with that.
At this crossroads, I think that if Motorola chooses the path of embedded only (and they don't push high-end embedded) then they will end up (over years) to just become another footnote in the silicon industry. They need to stay hungry -- but management leaving the mainstream market is basically a signal to me, that "it is just too hard to compete in the mainstream processor market" -- and that will pretty much mean that IBM will gain momentum in the mainstream, and they will use that momentum to help with embedded momentum, and Motorola is road-pizza. The mainstream has made Motorola competitive in the past, and will in the future -- and I think many in Motorola know this. They need the mainstream. It's not just about volume, it is about focus and direction.
But let's assume the worse, and Motorola blindly (and stupidly) takes the "embedded only" path (or at least embedded mostly) -- it still won't hurt Apple (at least in the short term). If Motorola pushes for more control of the embedded market, they will be doing high-end embedded (like AltiVec) -- which Apple can leverage. Remember, Motorola likes AltiVec because it is a good for signal processing (an Embedded like technology) -- so it isn't like "embedded market" means "old and slow". Worst case, the lower power and lower cost designs (more common with embedded market) would force Apple to push down into the less expensive appliance type computer market (iMacs, eMacs, etc.). This would increase volume, and certainly wouldn't mean the end of Apple or PowerPC.
This scenario is unlikely, since Apple still has IBM for the very high end, and Apple has the rights to produce PowerPC's on their own, and get other foundries to manufacture them for them as well.
But remember, the embedded market is fed by the mainstream. Motorola needs this to stay current. They need to push the boundaries, and Apple helps them do that. Apple is a pain to work with, but they are getting better -- and if Motorola has not walked away from Apple yet. Why should they? Apple is their biggest customer -- without Apple, I think Motorola semiconductor would have been sold off 10 or 15 years ago. Motorola needs Apple, and they will put up with Apple. And there is plenty of money between Apple and the embedded market to pay Motorola's R&D, even if they were to increase the money going into it. Now that IBM has gone out of Somersett, there can be more focus -- so the R&D money that is being spent can be spent more wisely. So it isn't like IBM breaking out of Somersett hurts Apple and Motorola, and in fact it may even help them. (This isn't meant to bash IBM, just the focus was different).
Whenever I read the chicken little's predicting that there is not enough money in the PowerPC marketplace to foster innovation I want to chuckle. I don't really think they are paying attention to the big picture.
Look at the PC marketplace -- AMD and Cyrix really don't have that big a piece of the pie. Intel does. Right now, it looks like AMD is making better processors, and more innovating designs than Intel. The K7 and K6-3 really have a good chance of taking lots of marketshare away from Intel. But how can they? They are small? They don't have enough R&D money to support their designs? It isn't like they can just borrow their designs from the rest of the x86 industry -- they have to design on their own. It isn't like their market share is bigger than the PowerPC market. So why does market logic only apply to x86 and not to PowerPC?
Now compared to AMD volumes, the DEC-ALPHA (nay, Compaq-Alpha), Sun-SPARC, and HP-PA architectures are teeny tiny. Yet they thrive on a fraction of AMD (and a fraction of Apple's Millions per year). So Apple could easily push to make their own chips too. There are some realities, like those other RISC designs are high end, and Apple is more mainstream, but the logistics aren't THAT different. Apple could definitely hold on long enough to shift to another architecture.
Because of all these facts (and opinion), I have no fear for Apple or the PowerPC. For now, it is serving Apple and the Macs well. I see a lot of reasons why it is in everyone's best interests (at least for a while) to make it continue to do so. If it suddenly isn't, or companies do stupid things (which we all know they've done before), Apple and the Mac will go on. But for now, there is plenty of money and motivation to make really hot new PowerPC's.
On the really positive side, there are a lot of things looking bad for Intel and the x86 architecture for now. This means more choice and divergence for the industry. I don't think Intel is going to pull off the Merced migration unscathed -- and it may be a big fat flop. IA64 / Merced is designed with the silly goal (IMHO) of stuffing more and more parallelism into a single thread of execution (at a time when we are seeing diminishing returns at doing this). At best I expect architectural performance increases of like 20-40% for Merced. I think multiple core parallelism is a far better idea, and likely to result in far bigger performance gains (say 100-300%) -- and having a nice small efficient core, like embedded PowerPC's require make this far easier to do. So I think Apple and Motorola's goals (and high-end processors with efficient embedded processors), are far more in line than people are giving them credit for.