The 7 Myths of the Cell Processor

[McFly]

http://games.slashdot.org/comments.pl?sid=138810&cid=11616545

# The Cell is just a PowerPC with some extra vector processing.
Not quite. The Cell is 9 complete yet simple CPU's in one. Each handles its own tasks with its own memory. Imagine 9 computers each with a really fast network connection to the other 8. You could problably treat them as extra vector processors, but you'd then miss out on a lot of potential applications. For instance, the small processors can talk to each other rather than work with the PowerPC at all.

# Sony will have to sell the PS3 at an incredible loss to make it competitive.
Hardly. Sony is following the same game plan as they did with their Emotion Engine in the PS2. Everyone thought that they were losing 1-200 bucks per machine at launch, but financial records have shown that besides the initial R&D (the cost of which is hard to figure out), they were only selling the PS2 at a small loss initially, and were breaking even by the end of the first year. By fabbing their own units, they took a huge risk, but they reaped huge benefits. Their risk and reward is roughly the same now as it was then.

# Apple is going to use this processor in their new machine.
Doubtful. The problem is that though the main CPU is PowerPC-based like current Apple chips, it is stripped down, and the Altivec support will be much lower than in current G5s. Unoptomized, Apple code would run like a G4 on this hardware. They would have to commit to a lot of R&D for their OS to use the additional 8 processors on the chip, and redesign all their tweaked Altivec code. It would not be a simple port. A couple of years to complete, at least.

# The parallel nature will make it impossible to program.
This is half-true. While it will be hard, most game logic will be performed on the traditional PowerPC part of the Cell, and thus normal to program. The difficult part will be concentrated in specific algorithms, like a physics engine, or certain AI. The modular nature of this code will mean that you could buy a physics engine already designed to fit into the 128k limitation of the subprocessor, and add the hooks into your code. Easy as pie.

# The Cell will do the graphics processing, leaving only rasterezation to the video card. Most likely false. The high-end video cards coming out now can process the rendering chain as fast as the Cell can, looking at the raw specs of 256Gflops from the Cell, as opposed to about 200GFlops from video cards. In two years, video cards will be capable of much more, and they are already optomized for this, where the Cell is not, so video cards will perform closer to the theoretical limits.

# The OS will handle the 8 additional vector processors so the programmer doesn't need to.
Bwahahaha! No way. This is a delicate bit of coding that is going to need to be tweaked by highly-paid coders for every single game. Letting on OS predictively determine what code needs to get sent to what processor to run is insane in this case. The cost of switching out instructions is going to be very high, so any switch will need to be carefully considered by the designer, or the frame-rate will hit rock-bottom.

# The Cell chip is too large to fab efficiently.
This is one myth that could be correct. The Cell is huge (relatively), and given IBM's problems in the recent past with making large, fast PowerPC chips, it's a huge gamble on the part of all parties involved that they can fab enough of these things.

Looks like he got it right ... but what do I know.

Fredi

 

[London Geezer]

Greed... Gluttony... Lust.........errr...... forgot.......

 

[McFly]

Here's an other interesting article: http://www.poughkeepsiejournal.com/today/frontpage/stories/fr020805s1.shtml

''This is still the biggest chip technology advance in probably 20 years,'' said Richard Doherty, research director at Envisioneering Group in Seaford, Nassau County.

If anything, claims of a 10-fold leap in performance are understated, Doherty said. ''Our estimate is 10 to 20, so they're being conservative,'' he said.

He added Cell developers said they could have put 16 cores on the same size chip if they had thought it necessary.

So I think we should not rule out a two PE or 16APU Cell chip already if Kutaragi is again willing to take some extra risks.

Fredi

 

[Megadrive1988]

Here's an other interesting article: http://www.poughkeepsiejournal.com/today/frontpage/stories/fr020805s1.shtml

''This is still the biggest chip technology advance in probably 20 years,'' said Richard Doherty, research director at Envisioneering Group in Seaford, Nassau County.

If anything, claims of a 10-fold leap in performance are understated, Doherty said. ''Our estimate is 10 to 20, so they're being conservative,'' he said.

He added Cell developers said they could have put 16 cores on the same size chip if they had thought it necessary.

So I think we should not rule out a two PE or 16APU Cell chip already if Kutaragi is again willing to take some extra risks.

Fredi

that would be nice. but only if PS3 has the RAM memory to make use of all that processing power.

 

[McFly]

that would be nice. but only if PS3 has the RAM memory to make use of all that processing power.

Myth #8: Without more than 256MB of RAM, more than 2 PE's are useless

More computing power is never useless!



Fredi

 

[marconelly!]

Wasn't it 256KB per SPU, not 128?

 

[Panajev2001a]

http://games.slashdot.org/comments.pl?sid=138810&cid=11616545

# The Cell is just a PowerPC with some extra vector processing.
Not quite. The Cell is 9 complete yet simple CPU's in one. Each handles its own tasks with its own memory. Imagine 9 computers each with a really fast network connection to the other 8. You could problably treat them as extra vector processors, but you'd then miss out on a lot of potential applications. For instance, the small processors can talk to each other rather than work with the PowerPC at all.

# Sony will have to sell the PS3 at an incredible loss to make it competitive.
Hardly. Sony is following the same game plan as they did with their Emotion Engine in the PS2. Everyone thought that they were losing 1-200 bucks per machine at launch, but financial records have shown that besides the initial R&D (the cost of which is hard to figure out), they were only selling the PS2 at a small loss initially, and were breaking even by the end of the first year. By fabbing their own units, they took a huge risk, but they reaped huge benefits. Their risk and reward is roughly the same now as it was then.

# Apple is going to use this processor in their new machine.
Doubtful. The problem is that though the main CPU is PowerPC-based like current Apple chips, it is stripped down, and the Altivec support will be much lower than in current G5s. Unoptomized, Apple code would run like a G4 on this hardware. They would have to commit to a lot of R&D for their OS to use the additional 8 processors on the chip, and redesign all their tweaked Altivec code. It would not be a simple port. A couple of years to complete, at least.

# The parallel nature will make it impossible to program.
This is half-true. While it will be hard, most game logic will be performed on the traditional PowerPC part of the Cell, and thus normal to program. The difficult part will be concentrated in specific algorithms, like a physics engine, or certain AI. The modular nature of this code will mean that you could buy a physics engine already designed to fit into the 128k limitation of the subprocessor, and add the hooks into your code. Easy as pie.

# The Cell will do the graphics processing, leaving only rasterezation to the video card. Most likely false. The high-end video cards coming out now can process the rendering chain as fast as the Cell can, looking at the raw specs of 256Gflops from the Cell, as opposed to about 200GFlops from video cards. In two years, video cards will be capable of much more, and they are already optomized for this, where the Cell is not, so video cards will perform closer to the theoretical limits.

# The OS will handle the 8 additional vector processors so the programmer doesn't need to.
Bwahahaha! No way. This is a delicate bit of coding that is going to need to be tweaked by highly-paid coders for every single game. Letting on OS predictively determine what code needs to get sent to what processor to run is insane in this case. The cost of switching out instructions is going to be very high, so any switch will need to be carefully considered by the designer, or the frame-rate will hit rock-bottom.

# The Cell chip is too large to fab efficiently.
This is one myth that could be correct. The Cell is huge (relatively), and given IBM's problems in the recent past with making large, fast PowerPC chips, it's a huge gamble on the part of all parties involved that they can fab enough of these things.

Looks like he got it right ... but what do I know.

Fredi

I disagree on the last point they make: the chip is even smaller than the Emotion Engine CPU that shipped with the PlayStation 2 in Japan and for which Sony was using 200 mm wafers while now the 90-65-45 nm chips will be realized using 300 mm wafers.
Not only the chips are smaller, but the wafers from which you obtain the chips have grown in area.

 

[PiNkY]

I disagree on the last point they make: the chip is even smaller than the Emotion Engine CPU that shipped with the PlayStation 2 in Japan and for which Sony was using 200 mm wafers while now the 90-65-45 nm chips will be realized using 300 mm wafers.
Not only the chips are smaller, but the wafers from which you obtain the chips have grown in area.

That reaoning implies that the difficulity of fabbing a respective 200mm^2 ic is about equal on a 250 micron and a 90 micron process. Also they seem much more aggresive in regards to clockspeed this time around.. Not that i'd have any intimate knowledge but based on intuition i'd say cell is a greater challenge....

 

[Panajev2001a]

I disagree on the last point they make: the chip is even smaller than the Emotion Engine CPU that shipped with the PlayStation 2 in Japan and for which Sony was using 200 mm wafers while now the 90-65-45 nm chips will be realized using 300 mm wafers.
Not only the chips are smaller, but the wafers from which you obtain the chips have grown in area.

That reaoning implies that the difficulity of fabbing a respective 200mm^2 ic is about equal on a 250 micron and a 90 micron process. Also they seem much more aggresive in regards to clockspeed this time around.. Not that i'd have any intimate knowledge but based on intuition i'd say cell is a greater challenge....

Well, I'd say that the experience of Toshiba and SCE has grown in the past 5 years and this time around they have IBM with them too.

The EE and the GS are not bad processor, technically if you think about it they were quite advanced for their time too: 6.2 GFLOPS (EE), 2.4 GB/s inner bus (EE), 3.2 GB/s Memory Interface thanks to embedded Direct RDRAM memory controller (EE), two independent and fully programmable Vector Units (EE), a 128 bits integer SIMD engine (EE), etc...

As far as problems with the 90 nm process, SCE and Toshiba have had time to put the process through practical tests as they have been fabbing the EE+GS@90 nm and the PSP SoC for quite a while (both use 90 nm technology).

 

[PiNkY]

Hmm, IBM's track record at 90nm is spotty at best. It was not too long ago that they weren't even able to meet apple's demand of ppc970(fx)s. And that's a design they basically have years of experience on. nVidia also relocated nv40 production to taiwan...

No point denying that EE/GS were nice chips, but at least the EE was comparatively lower clocked then other available cpu's at that time (Intel PIII (coppermine) was available at 800 MHz at Q4/99) with 28 million transistors compared to EE's 13,5...

 

[Mulciber]

am i totaly wrong, but isn't this chip being fabbed at a Sony plant, not IBM?

 

[V3]

Yeah, IBM going to fab Cell for their own purpose, from what I heard server and that joint workstation project.

Sony going to do the Cell for PS3, maybe with the help of Toshiba in the early stages.

 

[Panajev2001a]

Yeah, IBM going to fab Cell for their own purpose, from what I heard server and that joint workstation project.

Sony going to do the Cell for PS3, maybe with the help of Toshiba in the early stages.

SCE should have three sources for CELL chips:

1.) Oita #2 (co-owned with Toshiba)

2.) Nagasaki #2 (fully owned by SCE)

3.) East Fishkill's Sony/SCE+IBM plant (remember the $300 Million or so Sony/SCE invested in an next-generation manufacturing line with IBM ?).

 

[Megadrive1988]

Yeah, IBM going to fab Cell for their own purpose, from what I heard server and that joint workstation project.

Sony going to do the Cell for PS3, maybe with the help of Toshiba in the early stages.

SCE should have three sources for CELL chips:

1.) Oita #2 (co-owned with Toshiba)

2.) Nagasaki #2 (fully owned by SCE)

3.) East Fishkill's Sony/SCE+IBM plant (remember the $300 Million or so Sony/SCE invested in an next-generation manufacturing line with IBM ?).

that's nice. that's more sources than Nintendo will likely have for Revolution's GPU (NEC) and Microsoft will have for Xenon's VPU (TSMC)

 

[V3]

Perhaps East Fishkill is for the workstation Cell that Sony is working with IBM.

 

[Panajev2001a]

Perhaps East Fishkill is for the workstation Cell that Sony is working with IBM.

It will depend on how many chips the other two fabs can provide and how many chips SCE needs for PlayStation 3's launch in Japan, U.S.A. and Europe and how many chips per month they would like to produce (this relates to how many PlayStation 3 consoles they can produce per month).