CELL CHIP?? Big news for apple?

[GFLPraxis]

Bottom Line: Cell processor is likely to be the new MAC G6 (or whatever they call it. Perhaps it will be called MacCell.)

I don't know about you, but I just don't see this being called the G6. The Cell processor still has a PowerPC 970 inside with multiple APU's attached-it's technically a souped-up G5, and the very few programs that cannot take advantage of Cell (such as, say, VirtualPC) would run at the speed of the 4 GHz G5 processor in it.

I think this would be kept under the G5 name. G5 Extreme, perhaps?

HOWEVER, since the Cell can work with POWER4, POWER5, and PowerPC 970 (G5; based on POWER4), I think we can expect that when a POWER5-based PowerPC 980 comes out it will be called the G6 and it will go in the Cell as well.

 

[burger011]

who knew?

who knew that one day one of my post (other than the one where I was trying to figure out why my ipod shuffle wouldn't charge) would eventually be the first post in one of macrumors "headlined" discussions.

thanks macrumors!

burger

 

[GFLPraxis]

I would be surprised if Apple never released a Cell-based Macintosh, but it might not happen with the chip we see today. The number of APUs, from my understanding, is not set by the architecture, nor is the full nature of the core. I would expect to see a G4+2APU embedded chip, a G5+16APU supercomputer chip, and several variations in-between. In effect, you could think of the G5 with a different memory bus as a Cell with no APUs. I wouldn't be surprised to see the next generation PowerPC chip all be based on a Cell core.

Apple could offer the APU instructions as an option to developers, and integrate them into Quicktime and their own high-performance floating point and image manipulation libraries for noticable speedups to existing applications written to the APIs.

After 2 years of hearing how 64-bit "native" code will speed up software (it won't) and such, it's nice to see an innovation in the PowerPC field that really CAN speed up software.

Very reasonable. I doubt we will see a G4 with the APU's since it's made by Motorola and not IBM, or at the very least, it will come out at least a year later (Motorola might very well keep the G4 alive, they're pushing for 2 GHz this year...if they go 3 GHz when the G5 reaches 4.6 for the Cell, they could stay in low-end Macs).

But when a G6 comes out (a PowerPC chip based on POWER5), I'd expect it to work with Cell right away.

 

[relimw]

I think this would be kept under the G5 name. G5 Extreme, perhaps?

HOWEVER, since the Cell can work with POWER4, POWER5, and PowerPC 970 (G5; based on POWER4), I think we can expect that when a POWER5-based PowerPC 980 comes out it will be called the G6 and it will go in the Cell as well.

Or will they do like Intel did with the 586 (pentium)? Say, G5 II, G5 III, etc?

 

[rob_osx]

Why Apple will use the Cell

After reading many articles about the cell, here is why I think Apple will use the Cell processor.

• The Cell processor is based on the POWER architecture, which means it would be compatible with Apple’s software. Though some people disagree, stating “[the cell] processors are not compatible with Altivec” and “almost impossible to achieve backward compatibility”. This is totally wrong because I'm running Panther with a G3 that does not have an Altivec unit, and if you look at Apple's Core Image it seems as if they are heading away from the Altivec and using a GPU.

• The Cell processor is made to scale from low power, consumer electronics devices like the Playstation 3, to high performance requirements, servers. IBM has said that one rack of Cell servers would have around 16 teraflops of performance. Using Apple’s current Xserves it takes 40 racks to get 25 Tflop/s of performance. Apple would be silly to turn away from that type of performance/footprint ratio.

• The Cell processor will have a low cost, because of volume of production.

Go here if you want the details and references to the articles I used to support my ideas.

 

[GFLPraxis]

After reading many articles about the cell, here is why I think Apple will use the Cell processor.

• The Cell processor is based on the POWER architecture, which means it would be compatible with Apple’s software. Though some people disagree, stating “[the cell] processors are not compatible with Altivec” and “almost impossible to achieve backward compatibility”. This is totally wrong because I'm running Panther with a G3 that does not have an Altivec unit, and if you look at Apple's Core Image it seems as if they are heading away from the Altivec and using a GPU.

• The Cell processor is made to scale from low power, consumer electronics devices like the Playstation 3, to high performance requirements, servers. IBM has said that one rack of Cell servers would have around 16 teraflops of performance. Using Apple’s current Xserves it takes 40 racks to get 25 Tflop/s of performance. Apple would be silly to turn away from that type of performance/footprint ratio.

• The Cell processor will have a low cost, because of volume of production.

Go here if you want the details and references to the articles I used to support my ideas.

I need an appluading smiley. Excellent post.

Your second post works the other way, too. Due to its scalabilitiy, Apple could get the ultimate in performance out of their XServes...but they could also scale down, and put cell chips in something as small as the Mac Mini, as well.

 

[chaos86]

did nobody else notice this little idea:

the cell requires a HUGE amount of cooperation between hardware creators and software programmers. as alot of people have said, it will require a complete rewrite of any OS or software, and of course the hardware around it will be built from the ground up.

now apple, as the only computer company that makes both the hardware and the software, has a big advantage. the hardware and software teams can work side-by-side, whereas in the windows (for example) world, intel needs to make a prototype, let ms play with it and program it, intel revise it, ms revise the software, back and forth. and even then it will be full of bugs and problems for years with AMD and Intel both making systems and ms making their crap.

if all sides start at the same time, apple will win the initial race to incorporate the cell (properly).

 

[XForge]

Wow... This is a pretty big advancement, isn't it? We were having a discussion in our Adv. Computers class with the teacher (who was explaining the cell processor), and he said that the transistors in this thing are only a few atoms long Is that true? Or was he thinking about something else?

So electrons go through these transistors like bowling balls in the ball return?? = )

 

[Booga]

embedded to desktop

Very reasonable. I doubt we will see a G4 with the APU's since it's made by Motorola and not IBM, or at the very least, it will come out at least a year later (Motorola might very well keep the G4 alive, they're pushing for 2 GHz this year...if they go 3 GHz when the G5 reaches 4.6 for the Cell, they could stay in low-end Macs).

Pretend I said "G3+2APUs" could be embedded. IBM loved the G3, and would have preferred pumping it up the GHz rather than touching AltiVec, which they only did at Apple's insistence. Now we see why.

Remember, most CPUs sold today are NOT x86 CPUs. Last I checked the x86 hadn't broken 5% market share of TOTAL CPUs sold each year, and most CPUs sold today are still 8-bit. While the desktop market is currently very lopsided toward x86, there is a lot of money to be made in cars, DVD players, microwave ovens, security systems, whatever. If the Cell is really as scalable as they say, and IBM does a good job with selling cheap reference boards and supporting them well, we could easily see the Cell chip dwarf the x86 market in no time at all. As for desktop chips, hopefully economy of scale will keep it competitive in that arena, too.

 

[eric_townsend]

This would bolster the Mac line nicely

I think the perfect name for such a new PowerMac would be the QuantumMac...
I would think Apple would be foolish to not at least consider it... Depends on what the IBM/Sony/Toshiba consortium will offer Apple too.

 

[kcm]

The new multicode processor has already exceeded 4GHz in lab testing and is designed to power a variety of operating systems.

I don't think I saw this mentioned already -- should be multicore.

 

[mzlin]

More details here

My apologies if this has been posted already. Go to part 4 for some interesting speculation about Apple.

http://www.blachford.info/computer/Cells/Cell0.html

 

[Lord Blackadder]

Geez, I've spent some time reading to get to this end of the thread!

I've a notion:

If the prototype Cell contains an IBM Power5 on the die, and is clocking at 4+ GHZ already, doesn't that mean that IBM has the capability to make a 4+ GHz standalone PowerPC CPU? If not, why?

This seems to have even more relevence to Apple hardware than the Cell since a Power5 derivative is likely to replace the 970 eventually, IIRC.

 

[rob_osx]

did nobody else notice this little idea:

the cell requires a HUGE amount of cooperation between hardware creators and software programmers. as alot of people have said, it will require a complete rewrite of any OS or software, and of course the hardware around it will be built from the ground up.

I think the point you are missing is that to take FULL advantage of the Cell, some areas of software might have to be re-written. A new compiler would probably be able to handle some of this. How would you like a current G5, 970, running at 4.6GHz?! Sure software could run faster if optimized for the cell, but the fact that the processor is running almost twice as fast as the current 970 and the memory bus being faster, applications will run faster.

When Apple came out with the G4 with the Altivec, software was not optimized for that, but it still ran. Eventually software was tweaked to take advantage of the Altivec unit, the same will happen for the Cell.

I do agree that new hardware will have to be built to handle the Cell. This is mainly due to a new memory bus.

 

[rog]

When oh when is the PowerBook quad core Cell ever going to come out? Wintel laptops are running at 100GHz and they're only 5 bucks! Apple needs to catch up now!

 

[joeboy_45101]

I believe that software can be designed to emulate anything. How hard would it be to emulate the function of an x86 or PowerPC processor in a software program. I know that this would be considered extremely complex, but I don't think impossible. And if somebody could learn how to emulate specific hardware like an x86 processor then later on it would be easier to emulate different platforms. With a Cell type processor you could instruct certain cores to emulate an x86 processor running Windows and other cores to emulate a PowerPC processor running Mac OS X. If it could be done then it would be a tremedous step in system design. You could build a system using 1 or 2 Cell's maybe one to emulate a CPU and sound card and another to emulate the GPU and other pieces of harware.

 

[rob_osx]

I've a notion:

If the prototype Cell contains an IBM Power5 on the die, and is clocking at 4+ GHZ already, doesn't that mean that IBM has the capability to make a 4+ GHz standalone PowerPC CPU? If not, why?

This seems to have even more relevence to Apple hardware than the Cell since a Power5 derivative is likely to replace the 970 eventually, IIRC.

The 970 was not made to scale to the higher frequencies, and that is why we have not seen a 3.0+ GHz G5. You are correct in pointing out that the 3 or 4GHz barrier was not an IBM frequency barrier, but a barrier specific to the 970. The POWER5 can handle the higher frequencies and I belive this does prove IBM can create a higher frequency stand alone PPC CPU for Apple.

So the question is why hasn't Apple/IBM come out with a faster chip for the PowerMac. We know Steve wanted 3.0GHz the summer of 2004. IMHO IBM told Apple about the Cell, and instead of Apple giving money to IBM to build a new chip based off the POWER5, Apple decided to wait for the Cell. Yes, it would cost Apple some lost sales because they didn't have faster PowerMacs and possibly laptops in the short term, but given the amount of money it would take for a new chip Apple decided to wait for the Cell that they could put in laptops and servers. This is not just a small step forward, but a giant leap. Apple likes leaps, remember when the G5 came out and the commercials with tanks guarding it?

Apple wants a processor that has a future roadmap. The Cell has a roadmap for both high and low power requirements. This is what Apple wants a chip that can go into all their products, and one that they don't have to pay for all the R&D costs.

 

[Frobozz]

Here's a question. Could the Cell theoretically replace the graphics card? If it's so good with media, could OS X be modified to use Cell for OpenGL and all 3d calculations, meaning that Apple can remove the graphics card to make up for the extra cost of the cell processor?

Or is a dedicated graphics card still necessary?

(I know a considerable amount about processors, but not enough about GPUs)

That's a fascinating concept... but I wouldn't see why not. After all, Cell is designed for multimedia processing. Hmmmm....

 

[dontmatter]

Its much faster than a G5. Where did you find the 256 gigaflops number? Based on a news release in November, IBM stated: "The companies expect that a one rack Cell processor-based workstation will reach a performance of 16 teraflops or trillions of floating point calculations per second."

I expect that IBM will be using this in their future cluster supercomputers. I don't think that Apple will use the Cell, but they will use the PowerPC core from it, and they will likely benefit from the rest of the design.

equals how many gigaflops per processor? Meaning: How many processors are in "a one rack Cell processor-based workstation"

 

[MarkCollette]

Ok, here's a few facts from the articles, which my ramblings below are based on:
- Cell contains a PU, which i a fully functional Power5 derived chip, similar to what we on MacRumors called the G6 for the past year or so
- Cell also contains a bunch of APUs (8 in this chip) which are given little APU programs from the PU. These little programs run on the APUs, and do whatever calculations.
- The PU, being a fully functional Power5 derived CPU, has things like cache, SMT, Altivec, etc.
- The APUs are designed to crunch through streamed data, holding interim data in their registers, so it actually makes no sense to give them cache. Caches only make sense when you have to reaccess parts of memory, which you haven't stored in your registers.


Now my intuition/guesses:
- From a hardware point of view, Cell would require a full motherboard redesign compared to, say, the G5. But, they'll have had years of time to do it, by the time Cell chips ship
- From a software point of view, there are always two stages of supporting new hardware: (1) Make old software run on it (2) Take advantage of new features. So, I wouldn't expect new Cell Macs to come with some totally new operating system initially, but rather follow the pattern of the G5 with OS X 10.2.7.
- Taking the lead from Quartz Extreme and CoreImage and the Pixel Shader programming standards in OpenGL and DirectX, I think we're moving away from a programming model of having the CPU doing number crunching, towards more hardware accelleration.
- In the past, one used the CPU (Use the CPU's ALU and/or FPU), and then later one might use the SIMD unit, then came using the GPU, and soon we'll have the Cell's APU.
- There are several problems with having all of these options:

- Only some compilers properly support autovectorization, so right now we're taking advantage of SIMD units nearly as much as we could
- Not everyone has a GPU that supports Quartz Extreme or CoreImage, or Pixel Shader programs. So, one can't count on using the GPU, and even if one could, the tools and APIs for using it are even further behind support for SIMD units.
- We don't even have Cell's shipped, so who knows how long it will be before there are APIs to make use of the APUs?​

- At some point, with all of these different means of accellerated processing competing with each other for software support, the only real solution will be for regular software developers to code at a sufficiently high level, that the APIs or the operating system will then be able to, at install time or runtime, make the decision of which accelleration method to use.
- After a couple years, who knows, maybe one way wil win out and become standard, but for the next 5 years or so, things will get a lot more complex.
- So, look towards codecs and scientific programs, and compilers, and operating systems to all try to provide support for the new accellerations, while standard programs sit on the fence waiting for simple trivial support to be provided to them.
- Oh yes, and some people were wondering if Apple would use the Cell chips. My opinion is that when the Cell chips first ship, all of the capacity will probable go towards PS3, for at least 3-6 months. After that, I assume Apple will be able to get their hands on Cell chips in volume. If the don't support Cell, then I would expect share holders to sue the board for negligence.
- So, one question is, will Apple ship systems with what we called the G6 in the mean time, until Cell arrives, or will they just stall in the water until Cell? It might not be cost effective to make a new motherboard design for the G6, when another will be required for Cell. But, it'll probably be a year at least until Cell Macs come out, and IBM might not want to bother making more 970 chips as they focus more on Power5 and Cell.

 

[Catfish_Man]

The 970 was not made to scale to the higher frequencies, and that is why we have not seen a 3.0+ GHz G5. You are correct in pointing out that the 3 or 4GHz barrier was not an IBM frequency barrier, but a barrier specific to the 970. The POWER5 can handle the higher frequencies and I belive this does prove IBM can create a higher frequency stand alone PPC CPU for Apple.

So the question is why hasn't Apple/IBM come out with a faster chip for the PowerMac. We know Steve wanted 3.0GHz the summer of 2004. IMHO IBM told Apple about the Cell, and instead of Apple giving money to IBM to build a new chip based off the POWER5, Apple decided to wait for the Cell. Yes, it would cost Apple some lost sales because they didn't have faster PowerMacs and possibly laptops in the short term, but given the amount of money it would take for a new chip Apple decided to wait for the Cell that they could put in laptops and servers. This is not just a small step forward, but a giant leap. Apple likes leaps, remember when the G5 came out and the commercials with tanks guarding it?

Apple wants a processor that has a future roadmap. The Cell has a roadmap for both high and low power requirements. This is what Apple wants a chip that can go into all their products, and one that they don't have to pay for all the R&D costs.

Actually, the POWER5 tops out at 1.9GHz right now, with very good reason. The G5 goes a little faster than that, but not much. This new processor is NOT a G5 core, it's smaller and simpler, with less performance per clock, and longer pipelines so that it can reach higher clock frequencies. It also appears that the Xbox2 will be using the same core, but with 3 of the powerpc cores, instead of one PPC and 8 vector cores. It's not certain that the Xbox core is the same or very similar to the CELL's PPC core, but it seems extremely likely. I think we'll find that they are the same. I'd say that a design like the Xbox2 is *much* more practical for Apple, since the vector cores are not Altivec compatible, while the more general purpose core is. In other words, we're likely to see some sort of multicore 4GHz processor without the CELL's specialized vector coprocessors (note that they're not vector units like Altivec. They're separate cores, running separate threads or programs). Porting to CELL would be hard. Like, brain destroying, work for years hard. It's highly unlikely that CELL will be competitive without using its vector coprocessors, and those coprocessors will require a lot of rethinking of programs, and probably some very smart tools. The Xbox appears to be much more traditional; not as powerful, but much much easier to program for and get performance from.

 

[MarkCollette]

The 970 runs at higher MHz than the Power4, from which it was derived, because the Power4 is made to be more physically resilient, with thicker wiring, and includes more circuitry for error detection, etc. So, I fully expect the Power5 derivatives, used for the XBox and Cell, etc. to run at higher MHz than the Power5.

A lot of technical articles online describe the XBox and Cell chips, and I have seen nothing to indicate that they are the same. I believe they are similar, in that, at their core, they use Power5 derived technology, but I think they differ from there on.

In the very short term, it is likely that a computer with 3 Power5lite cores would be better than a Cell, in a desktop computer. But, as soon as Photoshop, FCP, iMovie, etc. all support the APUs, or better yet use a library that is retargetable to run on standard CPUs or APUs interchangeably, then Cell will prove itself the better choice. However it is, the outcome will be obvious within a year of their release.

Actually, the POWER5 tops out at 1.9GHz right now, with very good reason. The G5 goes a little faster than that, but not much. This new processor is NOT a G5 core, it's smaller and simpler, with less performance per clock, and longer pipelines so that it can reach higher clock frequencies. It also appears that the Xbox2 will be using the same core, but with 3 of the powerpc cores, instead of one PPC and 8 vector cores. It's not certain that the Xbox core is the same or very similar to the CELL's PPC core, but it seems extremely likely. I think we'll find that they are the same. I'd say that a design like the Xbox2 is *much* more practical for Apple, since the vector cores are not Altivec compatible, while the more general purpose core is. In other words, we're likely to see some sort of multicore 4GHz processor without the CELL's specialized vector coprocessors (note that they're not vector units like Altivec. They're separate cores, running separate threads or programs). Porting to CELL would be hard. Like, brain destroying, work for years hard. It's highly unlikely that CELL will be competitive without using its vector coprocessors, and those coprocessors will require a lot of rethinking of programs, and probably some very smart tools. The Xbox appears to be much more traditional; not as powerful, but much much easier to program for and get performance from.

 

[pubwvj]

iPod Video a.k.a. iPod Cell.

Yeah, finally, iPalm. A scaled down Mac for the shirt pocket.
PowerMac/iMac for the desktop that never moves
PowerBook/iBook for the rest of us
PowerPalm/iPalm for the rest of the time

Carry all your data, backed up continuously to your iPalm.
Real-World Ruggedness.

Incidentally, yes, the iPalm will also play music.

 

[pubwvj]

Which begs the question is this the CPU that will be powering Mac OS 11?

From a marketing point of view I question Apple ever using the moniker MacOS 11 because all too (two, to ) many people are confused by such things. They'll wonder if that is OS 11 or OS II or OS 2 (Gak!) and chaos will rain (er, reign ). That's a real marketing know-no unless you're doing it on purpose to play on it to beat your way through the skull and into the consumer's long term memory. This isn't such a case.

 

[pubwvj]

How hot does this thing run?

Yes, this was one of my first questions, power consumption. Is this a possible portable chip or not? Is it an EnergyStar or a EnergyGuzzler? Sipper or Slurper?

Hopefully they are designing the chip to scale it's power consumption. When you're plugged in and need the CRU it burns up to max temp. When you're on battery is slows and sips to make the juice last.