Advocacy
Dojo (HowTo)
Reference
Markets
Museum
News
Other
Myths
Press
General
Hack
Hardware
Interface
Software
Standards
People
Forensics
Web
CodeNames
Easter Eggs
History
Innovation
Sightings
Opinion
Martial Arts
ITIL
Thought
|
Many people have pointed out that the G5 is on Motorola's roadmap <http://www.mot.com/SPS/PowerPC/overview/newroadmap.pdf> and that there are discrepancies between what they say, and what I said. I appreciated the pointers, and corrections, but wanted to also clarify a few things.
Many times my articles are written weeks in advance -- and the site updates itself during the week (all automated). So many times, something like the G5 roadmap will come out on Tuesday, and my article will show up on Wednesday, and not reflect the "newer" information. So this is the why's. But I wouldn't change much if rewriting the article today -- and some of the differences between what I mentioned, and what Motorola shows on their roadmap should be addressed just to clarify and offer some perspective.
Roadmaps are not law
Motorola has had many things on their roadmaps that have not appeared. I remember the G3 was supposed to have a high-end version (604e like, better FPU and more execution units) and there was also going to be a version that had onboard L2 cache (or that was at least discussed). They didn't happen. We can ignore the whole PowerPC 620. I've also seen a few MHz targets missed, but others exceeded. This is not through malice or deceit on Motorola's part -- it is just that no plan survives confrontation with the enemy. And in these cases time, tactics, competition, and marketing plans (and customers changing purchase plans) all alter the plans (and are the enemy). So remember that a roadmap is a set of targets or goals, from the companies point of view at a given time.
Generations
Going from the G1 to G2 basically was a completely new core. I'm not sure if anything was the same. This migration was short (a couple years or less).
Going from G2 to G3 was a very mild change and optimizations -- tune this, tweak that, and add a different cache system -- but it took quite a while. The G2s were doing a good job, and so they kept getting process bumps and mild improvements (like the differences between a 603 and 603e, or 604 to 604e) -- so that generation had legs.
G3 to G4 migration was a bigger change (than G2->G3), with adding a new execution unit, changing the MP and FP systems, changing the size of the internal pipe, and so on. But it happened in a lot less time than the previous bump.
So what will be called a G5 and why? When will it happen? I think the answer is the G5 will come out (and be called a G5) when Motorola marketing wants it to. There are no hard and fast rules as to what is a generation of processor. And this applies to the x86 camp as well. The PentiumPro, PentiumII and PentiumIII all combined probably had less architectural change than just the G3-G4 change alone. So it is sort of arbitrary.
What is a G5?
Actually, I was quite pleased to read the roadmap and see how many similarities there were between what I was saying and what Motorola was saying. My speculation was based on very little real information, and lots of logic, reasoning and piecing together parts and concepts. They were hitting and adding all the same things that I mentioned -- just the order and naming seemed to vary. Obviously they know a lot more about their strategies than I do. I should have really titled my article "PowerPC Futures", and not G5, since modeling is pretty ambiguous.
Personally, I think if Motorola goes multicore, and adds L2 on board, that is more than enough change to call it a new generation -- even if the core is mostly the same, to users it would be dramatically differently. But most hardware guys won't consider it "new" until the core changes. So some tweaks to the core become the issue -- but how many? The roadmap implies the G4s are going to hit 1 GHz -- but I don't know how they will do that (even with adding SOI) in the near future -- at least not without some core tweaks. If they do tweak the core, then why shouldn't it be called the G5 to differentiate it from the current G4?
When a G5?
Some people are not expecting the next generation (G5) for another 2+ years. I don't think it will take that long for a few reasons.
Motorola does not usually hype a design concept until they are fairly close to releasing it. At Microprocessor Forum last year, Motorola talked about AltiVec -- they were already sampling the chips at the time -- less than a year later, they released the chip (and it could have been sooner). This year at the same show, they are discussing changes in the pipeline and what that will mean to performance. I don't think Motorola wants to pre-announce and give too much lead time for competition -- so it has a feel of being a year to a year and a half out. But, sure, Moto could be 2 years out on the new core -- but it is more likely that they want to release before then. Depending on how specific they get in the presentation will be a big hint as to how far out they really are.
Hype hurts. The PowerPCs are getting harmed by not having the same MHz as some other chips -- even if they are faster per cycle it just isn't as easy to market real performance (over perceived). Motorola and Apple know this. They want to get back up in MHz. So I think they will push this generation migration a little faster. The longer it takes, the worse off they are. So again, the normal 24 month cycle gets a little accellerated to a 12 to 18 months (probably closer to the latter -- and we come to the same conclusion.
Next year Intel is going to be releasing Merced (at least in samples) and hyping IA64. Soon after Merced is McKinley (a follow on to fix all the Merced flaws) -- probably less than a year and as little as 6 months behind Merced. Motorola needs something to counter that hype. They are in different markets (mostly), but it will still need a response (or one coming quickly on the heels of the other). So that again they have to have a response in the year to year and half time frame -- which still comes to the same conclusion.
Targets miss, and get slipped -- but I really think that they want 64 bit and new pipes in place as soon as possible -- and that means a new generation in less time than the normal 24 months. I bet they could hit 12 months if they really wanted to (and had been doing some development overlapped with the G4 release) -- but we'll all have to wait and see.
Process size?
Another thing that is getting ambiguous is process size. There are many ways to measure the size of transistors on a chip. So many variables and measures that it would require a rather large a complicated article, that I might write some day when I'm feeling very adventuresome. I'm not 100% sure, but I think Motorola recently changed how they measure a process (or what size they call things), to make things sound smaller.
I'm going to have to research this a bit -- but it seems like the predictions for the G4 were originally .22µ and even some of Apple and Motorola's documentation says .20µ -- but now they are selling it as .15µ. That is a very substantial jump (.18µ was in between), especially including the jump to copper. On top of that, we didn't see a dramatic change in performance, reduction of power and so on -- which we should expect to see from a jump to .15µ. So for now, I'm skeptical that the G4 is truly .15 -- or that the measurement means the same thing it did a little while ago -- and I doubt that Motorola just leapfrogged the industry. Usually IBM is better than Motorola in size (and performance) of process -- but Motorola out produces them.
I'd love to go into it more, and see what is really going on -- but I let my subscription to Microprocessor Report lapse (if they even detailed this issue), and it will take me some time to uncover what is really going on.
Conclusion
So I appreciate the feedback I got -- thanks. Hopefully, my clarifications at least help people understand why there are some ambiguities in what people are saying -- and where I am coming from. Don't believe everything you read -- from me or anyone else (including Motorola). Skepticism is a great tool -- don't ignore it.
Time marches quickly -- but especially quickly for processors. In fact there should be a warning, "Objects in the future may be closer than they appear!". And as I've said before -- the future is going to be exciting! Of course before I worry about the G5, I have to go out and buy myself one of those cool new G4 machines. I detect a cool Christmas present to myself coming!
|
