IBM dropping Cell

[Phynaz]

http://arstechnica.com/hardware/news/2009/11/end-of-the-line-for-ibms-cell.ars

 

[KingstonU]

This chip started out at 90nm then shrunk to 65nm and finally 45nm in the new PS3 Slim correct? Was there any enhancements to it other than die shrinks and lower cost?

 

[wwswimming]

i remember walking into the PC Club in San Diego in about 2002 and one of the store employees told me, "the Cell processor will make all of this obsolete", and he made a sweeping motion with his hand, indicating all the computers in the store.

i guess he was half right. PC Club went belly up about a year ago.

 

[Idontcare]

i remember walking into the PC Club in San Diego in about 2002 and one of the store employees told me, "the Cell processor will make all of this obsolete", and he made a sweeping motion with his hand, indicating all the computers in the store.

i guess he was half right. PC Club went belly up about a year ago.

lol wwsimming, that is the beauty of hype, it works which is why money is spent to do it.

This chip started out at 90nm then shrunk to 65nm and finally 45nm in the new PS3 Slim correct? Was there any enhancements to it other than die shrinks and lower cost?

Yep that is the progression:

Just in case you were curious, here's the similar life-cycle of shrinks for the PS2's brains:

 

[Cogman]

i remember walking into the PC Club in San Diego in about 2002 and one of the store employees told me, "the Cell processor will make all of this obsolete", and he made a sweeping motion with his hand, indicating all the computers in the store.

i guess he was half right. PC Club went belly up about a year ago.

lol, I've heard similar arguments of why Macs are eventually going to take out the PC market. Some BS along the lines of "Mac is the only company coming out with new technology."

 

[DrMrLordX]

The amount of hype that surrounded Cell was amazing. People were saying it would show up in every Sony product, from Playstation products on down to smartphones and what have you.

Any word on what Sony's position on future usage of Cell will be? They invested billions of dollars ramping up Cell production back in the day in a fit of spending that was staggering at the time. What's going to happen to all the fab capacity that was (or was meant to be) dedicated to production of current and future iterations of Cell? This is assuming, of course, that Sony will be incapable of future Cell development (die shrinks etc) and that they will be stuck with today's 45nm Cell.

 

[Arkaign]

This was sort of bound to happen unless Cell was used somewhere that could take advantage of added performance from die shrink/etc. With the shrinks to 65 and 45nm, it's highly likely that *much* greater clock speed / performance / cache size could have been implemented, but on a PS3, there would be no point. The games would all still have to run on the original Cell, and it would still run with that 7800 Nvidia variant, so ???

Multi-core is here to stay, however. When Cell was announced, the dominant PC processors in sales were still single-core.

 

[JFAMD]

It's too bad. It's not a bad processor, but so much of this business is about adoption and flexibility.

We work with a lot of embedded applications and most of them have flipped to x86 because of the widespread prevalence of the platforms and components. All of the electronic slot machines used to be proprietary years ago, but they have been moving to x86 because you can get economies of scale by spending your money on software instead of hardware and silicon.

 

[Idontcare]

Ironically enough the chief scientists for CELL (Peter Hofstee) foresaw in 2007 the impending collision his product would experience with AMD's Fusion (slide 23) and Nvidia's Cuda (slide 25).

http://www-05.ibm.com/hu/news/events/2007/bladenap/pdf/Hofstee_Cell.pdf

Software support is maturing and we have great momentum with the academic community due to our open software strategy​

Gap identified, apparently not resolved fast enough.

 

[waffleironhead]

So where were these available besides the PS3?

I remember mercury systems had an add-in board with a cell available for ~$8000. I am betting they didnt sell too many at that price tag.

 

[thilanliyan]

I think IBM sold some systems with Cell as well.

What kind of tasks was Cell actually good for?

 

[the kernel]

I think IBM sold some systems with Cell as well.

What kind of tasks was Cell actually good for?

There are a couple of areas where Cell was sold:

1) Toshiba sold a couple of laptop models with Cell (called the SpursEngine) which utilized Cell for some custom media software.

2) Leadtek and Thomas-Canopus sold add in boards based on the Toshiba SpursEngine that were drop in accelerators for media processing.

3) IBM has used the Cell processor (called PowerXCell) in several server implementations meant for the HPC crowd. It is used in a couple of supercomputer implementations (it can be combined with both POWER6 as well as traditional x86) and was most famously deployed in the "Roadrunner" supercomputer at Los Alamos (currently the #2 supercomputer in the world).

The problem with Cell is twofold:

1) Custom software needs to be written to harness it. This is a similar to the general GPGPU problem. This isn't so much an issue for the HPC crowd (HPC workloads are all custom written and hand tuned anyway), but it was a serious issue for the desktop/workstation crowd. The SpursEngine add-in boards are cool, but they are really just written to accelerate one or two rendering applications and as such are similar to the old hardware video boards that Matrox and Pinnacle used to sell to speed up Premiere rendering.

2) Cell doesn't solve any problems better than writing GPGPU applications using OpenCL/CUDA, and that's an issue because consumer machines already come with GPUs whereas Cell would be a chip that would have to be added purely for specialized computing.

In the end, IBM knows which way the wind is blowing and if anything it's designs like Fermi and Larrabee that are the final nail in the coffin for Cell.

EDIT: I think it's also interesting to look at WHY Cell exists in the first place, and for that we have to go back to the PS2. The PS2 design had a relatively unconventional design in that it mated a CPU with two powerful FP units (the Emotion Engine) to a dumb rendering chip (Graphics Synthesizer). The effect of this was that the heavy FP duties were done by the EE processor while the GS handled straight rendering duties.

This was at a time when tasks like T&L and pixel/vertex shading were being moved off of the CPU in favor of fixed function hardware on the graphics chip (starting with the GeForce and then later the GeForce3). Sony decided to go the other way with the PS2 which led to a number of disadvantages:

1) They needed insane levels of system Bandwidth to move all that data around between memory and the two processors. Because of this the PS2 was always bandwidth starved.

2) Technically the VPUs on the Emotion Engine could be used for things other than graphics (that was one of the chief "pros" in using such a design) but in reality the bandwidth limitations meant that although theoretically it was "programable", a modern GPU for the time could still produce better results. Mostly the only interesting things done with the VPUs was implementing things like realtime Dolby Digital/DTS encoding.

When Sony was doing the design for the PS3, instead of realizing that GPUs were becoming increasingly programmable, they decided to double down on their two chip strategy by introducing Cell (which in many ways is similar to the EE in the PS2). However, when they were coming into the later parts of the design cycle, Sony realized that a modern GPU would actually outperform the new Graphics Synthesizer design at a lower cost, so they swapped in the nVidia RSX chip. This left them with a terribly unbalanced design--after all if the nVidia GPU would do all the heavy FP lifting what was all that expensive Cell logic for? However they were already committed, and the PS3 shipped as the Frankenstein that it is.

Cell is an interesting solution to a number of problems, but Sony ended up financing a lot of its development for something that really didn't add any value to the PS3 platform. IBM may have found some cool uses for it, but Sony really got screwed royally by the deal. It was salt in the wound that Sony's R&D money also ended up financing the chip that went into the Xbox 360 and that product even shipped first to boot!