IBM cancels Cell Processor development

It has been confirmed that IBM will be pulling out of Cell development, with their current PowerXCell 8i to be the company’s last entrance in the technology.

The Cell Processor was originally the brain child of ‘The Father of PlayStation’ Ken Kutaragi. It was later co-developed by Sony, IBM and Toshiba; making its way into the PlayStation 3, TV’s and super computers.

Featuring a central general-purpose processor, the Cell includes a number of co-processing units with internal memory, known as SPUs. Greatly increasing performance, the Cell’s SPUs have been notoriously difficult to develop for, with a number of developers, such as Naughty Dog, making significant breakthroughs of late.

IBM’s most recent development in the Cell Processor’s design is the PoweXCell 8i, which is featured in the second most powerful supercomputer in the world; Roadrunner.

However, this is where Cell will end for IBM. The company’s Vice President David Turek, told German website Heise Online that the planned 32 SPE Cell processor will not be made.

However, Turek did explain that features of the Cell would continue to be moulded into other processor designs. With the future looking like it’s taking a GPU route, a hybrid technology is the direction IBM will developing.

So what does this mean for Sony and the PlayStation 4? It has long been understood that the platform holder wants to use the Cell processor once again, allowing them to take advantage of this generation’s research and to make the transition into the next much smoother.

Will this change with IBM pulling out of their own development? Not necessarily. Sony can still hire IBM to create a Cell Processor for their next console, without IBM being involved in their own internal development outside of the PS4. So don’t start counting Sony’s chickens just yet.

Readers Comments (20)

  1. If IBM is pulling out Cell then trying to develop powerful CPU+GPU hybrid then it’s good.
    PS3 is still in it’s 3rd year. IBM still has 7 years to develop a CPU+GPU hybrid for the PS4. There’s still a lot of time for IBM’s development.

  2. Well there’s no seven years, as development would have to be finalised around 2012 with a launch to be in the next couple of years. A 10 year life-scale doesn’t mean the new console won’t launch during that ten-years, just like the PS1 and PS2.

    Plus the problem with moving away from the Cell is backwards compatibility with PS3 games.

  3. I’ve always thought cell was an arcane architecture at best. Crunching bits means very little if the architecture is so unwieldy that no one can utilize the power of the processor. Any technology that relies on concurrent programming in order to take advantage of it is going to suffer, because concurrent programming is the one the most difficult things a programmer can do. Add that to the fact that programming is already one of the most difficult fields and you’ve got yourself a recipe for disaster.

  4. So what does this mean for Sony and the PlayStation 4? I dunno, a 16 SPU Cell? Three 8 SPU Cells? A Sony developed 32 SPU Cell? Something new made by IBM? Perhaps by some other company?
    Time will tell, and there’s no hurry. There’s a new xBox coming up, probably years earlier this time, I’m more interested in those specs. And other figures:) Meanwhile, in consoleland, ‘Cell exclusives’ are simply the best looking games with the bar being raised evey few months & no end in sight. MotorStorm, Heavenly Sword, Uncharted, Ratched&Clank, GT5:P, MGS4, WipEout HD, LBP, Killzone 2, Uncharted 2…God of War 3, Heavy Rain, The Last Guardian, GT5, Agent and that’s just the start, not bad for a 8 SPU Cell machine.

  5. Edgar: “No one can utilise the power” is now a moot argument. Naughty Dog has utilised the power, as has Sony Santa Monica, and many other teams. The PS2 was more difficult to develop for than the PS3’s Cell and it has had many advantages. Sure, it’s tough, but it’s seeing gains now.

    Plus, development won’t necessarily get easier – studio’s needed to be forced into developing for parallel processors, because it was coming sooner rather than later.

  6. Edgar, I don’t think you know what you’re talking about. The Cell is proven to be very powerful. It’s being used in scientific research to fight cancer, calculate black holes, by the military, by other labs to crunch numbers and render quicker. And Game Developers working with it exclusively are puting out amazing results. You are simply wrong.

  7. Edgar:

    Yes the Cell is based on old architecture – but that doesn’t mean that architectural design is flawed. It is in fact parallel processing – what supercomputers have been doing since the 70’s. Kutaragi’s vision was realized due to nano-scopic miniaturization that allows you to build a nine chip parallel processing architecture on a base of (i think it is now) 1.75cm*1.75cm.

    As Johnno correctly points out the Cell isn’t just used for the PS3 and it was never IBM’s intention that it should solely be for the PS3 – hence ‘Road Runner’ and the generation of ‘cell blade’ servers that IBM build.

    You are also incorrect to assume that parallel processing is nay more difficult than sequential coding. Often, sequential coding is put through a software program that transforms it into parallel processing scripting anyway (ie. it divides the scripting into lots of small packets to be distributed simultaneously to the 6 dedicated SPU’s in the PS3’s cell or the 8 * (no. of Cells on each blade * no. of blades) dedicated SPU’s in a Cell Blade server.).

    As we have seen with the recent exclusives (Uncharted 2, KillZone 2, etc.) – the PS3 is able to generate graphical output that the xbox360’s 3 threaded core sequential system just can’t match. Graphical output is usually measured by Flops (floating point calculations) per second. The xbox360 can generate 1 Tetraflop a sec. The PS3 can generate 2.2 Tetraflop per second (we are talking about the throughput of the complete system here – not the processors alone).

    As for the future – A hybrid cpu/gpu cell? That would certainly be welcome in a PS4!!

  8. Who the hell is waiting 10 years for a PS4? Freaks……………….

  9. Consoles keep the same hardware for a long time, while computers always change. IBM not supporting it anymore makes since. Actually I’m surprise they lasted this long!

  10. Patrick Sean: That’s two whole developers that have managed to master cell, and even then only with a lot of funding from Sony. Numerous other developers have complained about the complexities of developing for the cell processor. Given that most cross platform games actually look better on the supposedly weaker XBox 360 I think they have a good point. Who knows perhaps Sony will continue to make harnessing cell easier with better tools, but for now I personally remain unconvinced that Cell is a better route than more traditional architectures.

    Johno: The applications you bring up are scientific applications. Scientific applications tend to do their work through brute force calculations. What this means is they will attempt different permutations in order to solve a problem. Each permutation is usually a simple enough calculation, but the sheer number of permutations necessary to solve the problem makes finding the answer very computationally expensive. In those cases parrallel computing is actually very beneficial and fairly easy to implement in software. Obviously the cell processor is very well suited to these kinds of applications. Unfortunately we’re not talking about scientific applications. We’re talking about video games, and in many cases parallel computing just isn’t well suited to the problem. Often times you will need one result before you can move onto calculating another, and in those instances it becomes very difficult to use parallel computing. Synchronizing results is very difficult, I know I’ve had to do it a number of times in my own work. That being said there are definately certain parts of video games that lend themselves very well to parallel computing; such as graphics rendering. That’s why most video cards come packed with up to 128 small engine cores. Unfortunately the PS3’s cell processors can’t be easily utilized for graphics rendering. Instead the PS3’s cell processors are used for the business logic of the game, and business logic almost never lends itself to parallel computing. Full Disclosure: While I am a Software Engineer, I don’t work in the Video Game industry. I am restating what I’ve read when I say that cell is hard to utilize in graphics rendering, and I’m not speaking from personal experience. If a seasoned PS3 developer finds this to be false please correct me.

    Darth_Oblivion: Have you done parallel programming? I have. In fact I consider myself to be rather good at it, but that doesn’t mean I don’t recognize the complexities inherent in doing concurrent programming. You suggest that scripting languages can take care of this. It’s true that there are some declarative scripting languages like SQL that will actually take advantage of multiple processors when executing your statements without burdening the developer. Unfortunately declarative languages are typically only useful at very niche operations, i.e. SQL is very good at fetching data from a database. With video games you will need to use a much more powerful and general purpose programming language, typically C and C++. The need to remove the complexities of parallel computing has not been lost on companies like Microsoft. I am interested in Microsoft’s experimental programming language: Axiom. By the way when I said Cell was Arcane I didn’t mean it was old. I meant that Cell is difficult to harness. One more note FLOPS is a measurement of Floating Point Operations Per Second. Due to the cpu intensive nature of floating point operations they are rarely used in video game development. Typically video games use math libraries that use integer math, so no graphical output is not measured in FLOPS.

  11. clearly the people who keep mentioning that ps3 development is getting better have no clue at all about what it takes to make all those pretty things that they see in the naughty dog video games. The difference between Uncharted 1 and 2 is not in the programming but time and money spent moving little bits around the processors. Doing this is not beneficial in the long run (cause in order to get better results you had to keep spending more and more money and more and more time) especially when there are obviously easier ways of achieving the same result. The problem that the cell had is the its a one trick pony and it didn’t even do it well.

  12. Correction to my previous post. The incubation language at Microsoft that is intended for parallel computing is called Axum, not Axiom.

  13. Owen: What you’re saying is not actually correct. Uncharted cost around the same as any other video game developed, and due to ND’s strict development time, probably less – around $30million.

    What makes the difference at ND is their developers – they are some of the best in the industry, and this shows in the fact that the PS3 development team ICE is in-house with them (ie. ICE=Naughty Dog). ICE develop the EDGE tools for all PS3 developers, and so every breakthrough ND make on the Cell, they put into these tools.

    They’ve made a whole bunch of breakthroughs and moved much of their GPU work onto the SPUs. And this is the only way games like Killzone 2 and Uncharted 2 can look this good on the PS3, and better than on the 360. Since the PS3 has an underpowered GPU compared to the 360, the PS3’s only advantage is the Cell. So the fact that the PS3 can have better looking games even with a “crapper” GPU – shows that with the Cell, they can get great results.

    Though there will be better ways in the future, the Cell’s design and lessons will be something IBM will be implementing into new Chip designs for some time.

  14. Edgar: Just a point – “Unfortunately the PS3’s cell processors can’t be easily utilized for graphics rendering.” – this only isn’t wrong, it’s contradictory to reality.

    What you say now is what they thought in 2007. But since then, developers have been putting most of their GPU operations onto the Cell – something that no other Chip design has been able to do (at least in the console space) – and something that IBM will be continuing with hybrid designs.

  15. PS3 will be around for a very long time. For Xbox360 I believe they will have to come out with a Blueray version sooner or later and do some major engineering fixes regarding issues. We have still lot of lead way to play in before anything better is required for long time since the home video source is lock at 1080P. The question is even if a PS4 gets develop will the graphic quality be noticeable or that much better. HD now is 1080P and will be the standard for many many years. People are still in the process of upgrading there TV. We have reach a plateau when it comes to graphics and not go into the cost expenditure of high end GPU’s that are currently only available for PC’s in the 300 to 600 dollard range. Sony and Microsoft still have to recuperate much of the RND and loss from the sales of the current systems. I don’t believe we will see a new system in the next 4 years if not more.

  16. Just thought I’d mention that the waster gamer blog has copy pasted your post.

  17. Thanks John. Naughty.

  18. Patrick Steen July 18, 2010 @ 06:08

    The author is not mistaken, that actually only expands upon this statement within the article:

    “However, Turek did explain that features of the Cell would continue to be moulded into other processor designs. With the future looking like it’s taking a GPU route, a hybrid technology is the direction IBM will developing.”

    Cell is dead – the lessons from Cell will continue.

Comments are closed.