New Powermacs to use Cell Processor?
This mystery author, in a rather lengthy article, makes a case for the Cell Processor being the key to the advancement of the Powermac. I wonder what some of the more technically astute members of this forum might think about his ideas.
Article here.
Article here.
Comments
Originally posted by Yevgeny
NO. NEW POWERMACS WILL NOT USE THE CELL PROCESSOR. NEW POWERMACS WILL USE THE G5 CHIP. CELL IS WILDLY HYPED.
Well, I guess that settles it. Excuse me for breathing.
For one reason, code needs to be largely rewritten, not just recompiled.
For one reason, code needs to be largely rewritten, not just recompiled. [/B]
If that is true, then it would be easier to move to the Pentium than the Cell processor.
Why is everyone so hot on the Cell and down on the Pentium?
I know that 8088-family assembly language is crap, but that only affects the compiler writers...
The presenter talks about Core Image as a "stream based processing model" and "mappable to GPUs, SMT CPUs and any other parallell architectures ".
"Any other parallel architectures"? Like Cell, perhaps?
PS- Here are links to other speculative articles by "Neo"...
leap to MAN
iPod-Live
aacPlus
GarageBand-Live
Apple's Connexion to Boeing
Originally posted by e1618978
If that is true, then it would be easier to move to the Pentium than the Cell processor.
Not necessarily, since Cell contains a PowerPC, and the x86 lacks both a minimum binary compatibility and AltiVec or anything comparable. SSE and clones aren't even in the same league.
Why is everyone so hot on the Cell and down on the Pentium?
Because x86 is just the same basic style as the traditional PowerPC, only more poorly designed and more fragmented.
Cell changes the rules.
More to the point, Cell uses parallel SIMD processing, um, cells, which make it a very promising multimedia architecture. The Pentium offers all the pain of a transition to a wholly incompatible ISA, and no compelling features over a standard PowerPC core. (A fairly minor, transient speed differential doesn't count—there have been times when PowerPCs were faster, too). It would require a lot of work to get OS X to really use a Cell architecture, but at least the work would pay off with capabilities that conventional CPUs simply don't have.
I know that 8088-family assembly language is crap, but that only affects the compiler writers...
The endian issue affects everything, the lack of AltiVec severely reduces the functionality of some of Apple's basic apps and technologies, the lack of binary compatibility forces apps to be rebuilt to get any kind of performance at all, and the completely different design priorities require every app to be completely reoptimized—and possibly re-architected—to perform well.
And again, this is true to some extent for Cell, but at least Cell gives us something we didn't have before, in the form of multiple, superfast SIMD cores. Also, a move to the Pentium would be a move back toward an architecture that still prefers big, single threads (even if there can be two of them now) and dual processing at best. Cell encourages multithreading and small, dedicated tasks—which Cocoa also encourages—and like PowerPCs generally, is much more MP friendly.
Having said all that, the most obvious upgrade path for the PowerMac is the 900-series PowerPC. The Power5-derived member of that family should answer any performance objections (its big brother, the Power5, certainly has) and the total amount of work required to get OS X and OS X applications running well is just about zero—it's a straight-up PowerPC. Getting the kernel to exploit FastPath, if that makes it down from the Power5, shouldn't take too much work on the part of the kernel team, and no code outside the kernel should know or care whether FastPath is there.
Originally posted by e1618978
If that is true, then it would be easier to move to the Pentium than the Cell processor.
Why is everyone so hot on the Cell and down on the Pentium?
I know that 8088-family assembly language is crap, but that only affects the compiler writers...
Because the x86 is an architecture that consumes a lot of electrical power to reach the amount of computational power that it does. That is, it is inefficient. It also has the dumbest FPU implementation ever, and a hugely superfluous instruction set. Even the Penitum-M is inefficient. It just knows when to curb the clock.
The current state of the Pentium is purely a manifestation of Microsoft's habit of releasing new OS's for the X86 that require more and more overhead, in conjunction with the consumer fixation that higher clock speed is better. Intel wishes the X86 were dead. Upgrading it siphons off a lot of good engineers that should be working on more promising architectures.
Otherwise, Amorph covers the topic well.
Originally posted by Splinemodel
Intel wishes the X86 were dead.
Ah, yes, thanks for reminding me of that detail. Yes, Virginia, Intel has been trying to kill x86 for the last decade.
It's hard to do when the world's most entrenched operating system depends on it, though.
Linky
Linky
I hope Apple ends up using it in some way if it gives their machines a big performance boost.
Should be interesting next week at the International Solid State Circuits Conference.
Session 10.2: The Design and Implementation of a First-Generation CELL Processor
A CELL Processor is a multi-core chip consisting of a 64b Power architecture processor, multiple streaming processors, a flexible IO interface, and a memory interface controller. This SoC is implemented in 90nm SOI technology. The chip is designed with a high degree of modularity and reuse to maximize the custom circuit content and achieve a high-frequency clock-rate.
Category: "Things that make you go Hmmmm..."
BadAndy at Ars also followed that line of thought, apparently with a little help from a friend (way to call him out, too): The real appeal of Cell is that it's not just built around a 64 bit PPC—er, excuse me, Power architecture processor (gee, like what?), it's a modular architecture. So Microsoft can get one thing for their Xbox, and Sony can get something else for the PS3, and Apple can get something else, and IBM takes it all to the bank.
It's like Book E for performance mavens.
Hmm indeed.
Originally posted by Amorph
So Microsoft can get one thing for their Xbox...
Would Sony really license the tech to a rival (I assume it is jointly designed)?
Everything I have heard until this point is that Xbox2 will not use Cell (just a customised PPC). If this is the case, where do the patents exist for IBM, say, to design a similar core for M$?
(I really don't know with jointly-engineered projects like this.)
[Edit: Just re-read your post. Sorry, I was assuming you werre talking about Cell specifically rather than IBMs use of Power.]
Originally posted by Programmer
Category: "Things that make you go Hmmmm..."
Cell is going to be built on a 90 nm process? Isn't that going to be a rather power hungry processor judging from the 970fx? Hmmm...
Could someone help this poor ignorant bystander and provide a better explanation, please?
I feel like I'm perpetually in a state of confusion.
Originally posted by DaveLee
Would Sony really license the tech to a rival (I assume it is jointly designed)?
Everything I have heard until this point is that Xbox2 will not use Cell (just a customised PPC). If this is the case, where do the patents exist for IBM, say, to design a similar core for M$?
(I really don't know with jointly-engineered projects like this.)
[Edit: Just re-read your post. Sorry, I was assuming you werre talking about Cell specifically rather than IBMs use of Power.]
BA's point, over at Ars, is that the Power core looks like it is common between the two. This part is clearly IBM IP. The fact that SMT is talked about implies some kind of a relationship to POWER5, as does the timing (a year after POWER5 arrived). The POWER5's emphasis on low power consumption also fits nicely.
Cell is a whole lot more than the Power core.
Cell is going to be built on a 90 nm process? Isn't that going to be a rather power hungry processor judging from the 970fx?
You can't judge one processor by another. There are many design choices and trade offs made in every processor. Plus IBM has some tricks it hasn't yet rolled out into the 970 line of processors, at the 90 nm node.
Originally posted by rickag
What's with all the "hmm's"?
hmm=Home Media Mac
hmmm= Home Media Mac Mini
(Just to confuse you more. Maybe that way someone will be even more confused than I.)