Is Cell a Threat?
http://www.macnewsworld.com/story/38633.html
This article was linked in the Power5 thread, but I believe it deserves a separate discussion. The author, Paul Murphy, paints a dark future for Apple at the hand of IBM, and I find it difficult to believe it could be true. His opinions are so different from anything I've heard so far. Yet he appears to know a lot. Is this doom and gloom just nonsense and full of logical holes? I certainly hope so.
I don't know enough to contribute a lot, but have notice a couple oversights. He says Sun is Apple's only hope for the future, but fails to address how Alti-Vec would be accommodated. Also, he does not mention that IBM is in the independent chip making business, and Apple could still contract chips from them, even if IBM were heading in a different direction.
Comments? I hope Programmer reads this. I know a developer personally, but I cannot disturb him now with such questions. He's getting a product ready for beta and working almost around the clock.
Jerry
This article was linked in the Power5 thread, but I believe it deserves a separate discussion. The author, Paul Murphy, paints a dark future for Apple at the hand of IBM, and I find it difficult to believe it could be true. His opinions are so different from anything I've heard so far. Yet he appears to know a lot. Is this doom and gloom just nonsense and full of logical holes? I certainly hope so.
I don't know enough to contribute a lot, but have notice a couple oversights. He says Sun is Apple's only hope for the future, but fails to address how Alti-Vec would be accommodated. Also, he does not mention that IBM is in the independent chip making business, and Apple could still contract chips from them, even if IBM were heading in a different direction.
Comments? I hope Programmer reads this. I know a developer personally, but I cannot disturb him now with such questions. He's getting a product ready for beta and working almost around the clock.
Jerry
Comments
Originally posted by quagmire
I swear this guy did the same article a couple months back. I remember reading this word for word back in September or October. If true this guy is desperate for news.
I believe you are right. In looking back to their home page, this listing has "Best of ECT News" on it, meaning it's a re-run I think. Likely it is just fill for the holiday, while nobody is writing. They could at least put the correct date on it so it doesn't look new.
Still, it is a disturbing read, even if old. I hope he's way off.
If, as indicated, the cell processor will be fast, cheap, and flexible it may well be that it is time for Apple to ditch Altivec and move along. This might well be part of the reason that the IBM/Apple discussion have been somewhat confusing about exactly what they would do together...it is not yet decided. With a multi-core, partitionable CPU with a clear price/performance advantage (and a prospect of long term development) it would be worthwhile to make the operating system changes necessary. IBM has the cash to do that if Apple does not and it seems clear that IBM likes what Apple has been doing with the X-Serve.
The article also talks about the lack of "management bandwidth" in a Sun/Apple combination to pull off the change. The fact is that any sort of acquisition of Apple will have to be done in such a way as to integrate the essential elements of Apple (and Steve) into the offspring or loose much of the reason for the acquisition in the first place.
My bet is on IBM, not Sun.
Originally posted by RBR
If, as indicated, the cell processor will be fast, cheap, and flexible it may well be that it is time for Apple to ditch Altivec and move along.
That would be insane, as AltiVec is just starting to really take off. IBM is about to roll out a compiler with autovectorization. Strange as it sounds, that particular innovation is just starting to hit its stride.
AltiVec is 2 additional processing units integrated into a PPC core. The implementation on the 7445 takes up less than 10% of the die. It's not a big deal to include it anymore.
Cell is a PPC core + a bunch of simple, high-speed SIMD units. Since AltiVec is an optional and relatively small part of a PPC core, it's an optional part of a Cell processor. Since it has strengths that the independent Cell cores will not, and the independent Cell cores will have strengths that AltiVec does not, it makes perfect sense to roll out an Cell CPU built around a PPC core with AltiVec.
Such a thing would change the Mac landscape considerably, in fact... not in a bad way.
Crap then, still crap now.
Let's see, a low-power, rapid fast new PPC-based multi-core design with the option of AltiVec included quite trivially...
Naw, I can't see Apple having *any* need for *that*.
The core PPC architecture and instruction set hasn't changed since its inception, which means that a compiler for a PowerPC 970 needs some tweaking for a Cell to be sure, but it's nothing like the shift from say a P4 to a Centrino, or worse, Itanium.
64-bit? Check. Added to gcc in a few months. Didn't require a whole new compiler, just a new module.
AltiVec? Ditto.
Multi-CPU? Way done.
Multi-core? Ditto.
A new PPC derived CPU doesn't mean having to start from more or less scratch (or worse, support 20 year old technologies) like it does over on the other side. Different rules apply here, and the analysts who have just a mild understanding of technology in the first place don't seem to understand that. They never have.
Originally posted by snoopy
Comments? I hope Programmer reads this.
Don't worry about it. Just another writer pontificating without any real information or domain knowledge.
I thought it was fairly well known by now that Cell will be using a PPC 970 series unit to coordinate the subordinate processing units, anyway. So if it has that, I don't see why it is so unimagineable for VMX (Altivec ala IBM) to exist on the die, as well. This is entirely ignoring that all those existing banks of SIMD style sub-units on a Cell die couldn't be any more well suited to emulate VMX outright. In a most optimistic scenario, Cell might have the means to emulate what is essentially 512-bit or 1024-bit Altivec engines?! (4 and 5x wide Altivec processing is what I'm getting at) Now don't tell me Apple couldn't use something like that... Can you say, "realtime filters in Photoshop, that you tweak by simply dragging a screen slider"? Ok, who knows if any of that is possible. I guess the more important note is Apple couldn't have a better CPU partner right now than IBM, wrt future-proof architecture/technology. Jump to Sun processors?! Is this guy on crack?
Originally posted by Randycat99
I thought it was fairly well known by now that Cell will be using a PPC 970 series unit to coordinate the subordinate processing units, anyway.
Considering nothing has been published, nothing is "fairly well known". There are just a lot of guesses and suppositions.
Originally posted by Randycat99
Well, that's why I said "fairly well known", not "absolutely well known".
Lets try "broadly speculated" then. To me the word "known" is too strong for this situation.
The major point is, it is a partnership with IBM, hence the use of PPC's in some way, shape, or form, is not at all unlikely. If there can be a PPC, then the presence of Altivec is equally plausible and feasible. The current line of speculation right now is that the "housekeeping" CPU will be some sort of PPC 970 variant. Whether or not that comes to be true or not is largely irrelevant as to how plausible Altivec could appear or be emulated in a Cell architecture.
Your logic is impeccable. The major issue, in my mind, is how the IP is distributed across the IBM, Sony, Toshiba partnership.
That Altivec or its functionality could not be incorporated into a Cell architecture seems to be the dealbreaker premise the guy in this article was trying to push as a reason IBM could "hypothetically" be a dead end for Apple. He doesn't give any reasons for why not, but somehow a jump to an entirely different RISC architecture (Sun) seems beneficial to him (???).
Right. Like I said, the author of that article is clueless. Probably has shares in Sun or something.