End of Cell for IBM

There must've been silicon budgets, which is what he's talking about.

 

They already have , at least , one ; Team ICE

If I'm not mistaken they helped even first-gen PS3 games' development.

 

He referred to the ICE Team earlier in the thread, and it doesn't negate at all the point being made. Which for my part was that for the cost of one of the early effort 'exclusive IPs,' putting money towards easing developer access across the broader industry would have been money better spent.

 

yeah but is it going to be to the extent of powerful cpu/weak gpu vs powerful gpu/weak cpu? I mean having the cell didnt exactly cause a compromise on the rsx (just poorly planned?) The gpu can still be the equal of whatever other console hardware is there, whether or not a cell cpu is included

I don't expect the gpu to start doing cpu tasks in the next generation, at least not in a significant way. we'll need all the graphics processing power we can get.

 

And various shades in between.

If the other console has a larger GPU, you aren't going to be its equal unless their design sucks and yours rocks. That is, if 100 mm^2 of GPU can achieve the same as 150 mm^2 of a different GPU, that second GPU would be a bad choice.

The only other option is to spend more and fit in more silicon, where you can have both a better GPU and better CPU, but then you have a costlier machine.

Again, it all depends what silicon is put in the boxes. Let's take a game that has to solve a 3D fluid. We could have 200 mm^2 of silicon in our console. We could put in 100 mm^2 of CPU and that could be man enough for the task, leaving 100 mm^2 of silicon for graphics rendering. Now if instead we went with a 20:180 mm^2 split CPU:GPU, the CPU won't actually be able to do the job, so we have to eat into our 'graphics' processor's time (graphics in inverted commas because this processor is just crunching numbers, and they don't have to be related to creating pretty pictures). In this GPU-heavy design, we'll have to take some GPU cycles. If we use 45% of GPU time to help calculate the water physics, that'd be equivalent to '80 mm^2' as it were. If we could assign shaders to the job, it might actually be measurable in terms of die space, but that's not a usual metric!

Point is, whatever is in the box has to be used to solve the jobs the devs want. If they want 3D water and the CPU isn't up to it and the GPU can do the job, they are free to acquire GPU resources for their tasks. Taking an example, Pixeljunk Shooter has some great 2D physics. That's processing that could instead have been spent rendering a photorealistic game. The developers just decided the physics were more important than the visuals in this title and used the available processing resources accordingly.

For all the complexities of system design, it's actually all a pretty straightforward equation that's being balanced. You have execution units:data flow:flexibility, spread across whatever processors you have. You can reduce one to favour another, but could produce a completely imbalanced system. eg. All execution units and no bandwidth means no data to feed the execution units and they're a waste of silicon. All execution units and no flexibilityi means they can't do the jobs you need, or need to go around the houses to achieve simple tasks, wasting their performance. Keep the flexibility high and the cores increase in size, meaning the number of execution units per unit of silicon has to be decreased (for a given budget). For years, a split budget has shared the processing requirements across two cores: one highly flexible, low throughput CPU; and one inflexible, high throughput GPU. We're now starting to see mid-way processors that offer high throughput and reasonable flexibility, or higher flexibility with a little less maximum calculation power. These make scalable systems much easier to design, so are a logical advance for microprocessor design.

Oh, I should also add that developer effort is a factor in the system design equation, with a lesser effect on silicon budgets, but which can't go overlooked. Especially when you've got competitors offering developers an easy ride!

 

Being that I'm a bit of a latecomer to the discussion I guess I might just be plopping down a question on a train of thought we've passed, but such is life and here I go anyway.

On the discussion of investment and cost of support, development and research related to Cell I just have to ask if anyone here knows how cost effective the architecture is for Sony? They've sold their share in the Cell manufacturing plant to Toshiba, they haven't announced any new products containing Cell and now these rumors. For Sony to stick with it in the long run they'll have to make enough of a profit on it, so can Cell do that for them? If not I can really see Sony having a valid reason for changing to a more "mainstream" architecture as laymen (and others) here suggest/imagine. I kinda look at Cell with a sort of underdog mentality more then anything else, I want to to succeed, but it will have to go if Sony decide they can't make enough money on it in the long run.

 

Keeping Cell isn't about profit they might make, it's about total outlay vs an alternative solution going into the next product cycle.

 

Someone asked earlier in this thread or another one what all the millions Sony invested in the Cell has given in return. One thing I have concluded from the memory architecture they choose is that it is a high performing design that will be able do go through many steps of cost reductions through die shrinks. As the XDR memories get faster, they will soon be able to cut the number of data pins by 50% and thus make room for die shrinks to 28 nm and possibly below. We should also remember that Sony has Toshiba as a partner, who has started to use Cell in some hi-end TVs with an amazing feature set. Once Cell hit 32 nm and below Toshiba may migrate it to main stream TVs and economics of scale may kick in and help driving down Cell manufacturing costs.

The overall design of the PS3 is arguably the currently most forward looking console design, very much thanks to the BD player and the HDMI port. 5-6 years ago when Kutaragi put his final stamp of approval of the PS3 design, you bet he knew that 3D television was coming down the road and the PS3 should have all the capabilities to support that. If that was a wise decision businesswise can be discussed, but as an engineer I like technology that push the edge. There are probably more surprises down the road. I personally wouldn´t be surprised if the "true" next gen console from Sony is more than 5 years down the road.

When Kaz Hirai is saying that if the PS3 was easy to develop for it wouldn´t have a 10 year life cycle I think he really means it. I can´t understand that people get offended by it as it´s the nature of the console business, they even mentioned it the recent "PS3 development for Dummies" presentation.

It´s not hard to understand that the design of Cell took a lot of heat when it was introduced. It was well ahead of the curve when it came out with 9 separate cores in total in a heterogenous design. Now that game engines are starting to settle and get more stable Cell doesn´t seem to be that much of an issue any longer. On the contrary developers are starting to take advantage of it for advanced post-processing and such, where a lot of possibilities are left to explore.

Talking about hidden powers, I still don´t rule out that we will see a frequency bump of the RSX and the GDDR3 (currently down-clocked) memory within a few years, it would be in line with what Sony did with the PSP and follow their philosophy that "games have to get better every year". So when a lot of games are getting close to maxing out the PS3 capabilities they can give the GPU a 10% boost and let the games keep improving another few years. And of course using 100% of the Cell and RSX does not mean you are maxing out the PS3 capabilities as everyone knows, anyone can max out a CPU with a few lines of inefficient code. Sony has slowly released more and more of the reserved memory to the developers and that behaviour does also mimic the treatment of the PSP developers, it´s just a way to harness the overall development of games from year to year. Providing better toolsets is another way.

 

http://kotaku.com/5435384/report-so...=feed&utm_campaign=Feed:+kotaku/full+(Kotaku)

 

I think you basically condensed my post, which is awesome; But I believe that profit (or amount of profit) is decided on a investment-future development-profit analysis. If Sony doesn't see the potential future profit today that they saw when the initially launched the Cell architecture it will be replaced by something that does. My question still stands as to whether or not Sony still see this profit potential in Cell; Playstation isn't the "end all, be all" application for it.

I'm not against Cell, nor for a different architecture, this sort of analysis is made in most decisions ever made so I'm really just wondering if it still holds up, the same question you'd ask of any investment.

 

I am not really sure what you are asking or if I can answer it.

But anyway. Sony designed Cell and the PS3 with a very long life cycle in mind, which means it will be able to go through many cost-reduction and eventually probably end up with a cheaper BOM than what the PS2 currently have. Yes I am not kidding, I think so.

But that does not guarantee they will make profit from it. It all depends on that they sell enough qauntities. Do Sony think they will do that? Probably, but I don´t know what long term sales they are expecting. However, they have invested quite a lot in the Cell architecture through software tools beside the hardware itself so you bet they will do their best to take advantage of those investements. Swapping hardware now would would nullify quite a lot of those investments and increase hardware costs at the same time.

 

What would you like me to clarify?

 

Well, do you think anyone except Sony can give you a relevant answer to your question?

I doubt anyone on this board has access to any relevant numbers to make any analysis beyond the finger in air type of analysis I provided in my previous post.

 

Hehe, true enough. But I suspect there a few on here with a general (or assumed) knowledge of the profit generated, should that fall within a similar profit margin that other CPUs or related technology create then Sony has no direct incentive to change architecture.

Speculation is fun but not very constructive in most cases. I guess my posts sort of highlights that, sadly.

 

Cool! Cell<->PS4 still lives....probably a chance! Kaz please, at least, consult Ken!

What is so bad about using a Cell2 with upgraded LS thingy? Will it cost Sony a lot more money to switch some transistors? I would think even a Cell1 with upgraded PPC core @ 4Ghz+ @ 22nm @ 8 SPU is not a bad thing if you coupled it with the latest custom 22nm GPU from ATI in 2012, with the full FlexIO Cell-GPU bandwidth, and 4GB of 256bit GDDR5? At the least, BC would be a sure thing, programming the Cell is going to be old hat! All these would not be too expensive in 2012?

What would be the advantage of switching to a "multi-core PC" CPU, breaking BC and programming model, when isnt the GPU going to do the most "gaming" work anyway?

 

Color me Dan I wasn't condensing your post - that was the actual answer I was giving.

For PS4 they will simply choose the part that on a macro level makes the most sense economically. They have made a lot of investments into Cell, so, Cell in my mind is certainly a contender there. But as I and others have mentioned before, if another architecture were to offer as simple or simpler a development environment, similar performance characteristics when taken in the whole of the system, and cost as much or less than the Cell on a per unit basis... suddenly you have a viable alternative for the PS4 CPU that may be considered as its replacement.

I think the Cell as an architecture targeted towards a mass presence across several sectors and as a profit vehicle in its own right has honestly been abandoned by Sony, so when we talk Cell and Sony, in my mind we are talking only of the role it may or may not play in a future console.

 

Crossbar this line of thinking doesn't make much sense. If someone says that console games need to look better every year, that is not the same as saying that the architecture needs to launch with a built-in, purposefully steep learning curve. Games can show improvements and developers extract more performance when things don't punch them in the face as well, y'know?

Case in point - do you think that Sony will *purposefully* build in an equivalent learning curve into PS4? No, probably not... I would hope that anyone would recognize that. Because it would be ridiculous; you don't purposefully build in flaws into your system if you can help it, lest your grip on sanity be in question. On the contrary, it is clear they want PS4 to be 'developer friendly,' insomuch as possible. Believe me come the PS4 launch, you'll be reading quotes from Kaz extolling the greater approachability of the system.

I'm also of the mind that PS3 will never have a BOM lower than PS2's - either on an absolute or a relative basis. Firstly the HDD inclusion alone prevents it, but even assuming they go to a remote storage model in PS3v3.0, unless someone does a redesign of the RSX and/or Cell, I don't see them ever sharing the same die space... which as we know is a major factor in the PS2's path to cost reduction.

 

Can anyone tell me how much more you can do with spec A over spec B,

Spec A
Core i7 3.2ghz + 5870 + PCIE 2.0/3.0

Spec B
Cell 8SPU 3.2ghz + 5870 + FlexIO half/full rate

Isnt the problem of PS3 actually comes from the underpowered "last-gen" RSX?

In 2012, Cell1 ISA, even with upgrade PPU, will be very old behind the latest from Intel, however i dont think Sony will need to have the most powerful processors, but more on the total experience (hint: motion sensing, voice , facial recognition, 3d, augment reality etc), so i how about keep the "old" but "cheap" Cell1, even with PPU and SPU tweaks, add in "boatloads" of RAM 4-6GB. I wonder how much RAM can balance slower processors, and work better for a multitasking environment. Another savings for Kaz, comes from those motion reality Cell software, already shown to work and commercially available today. God damnit, somebody send my bright ideas to Kaz!

 

I re-read your post, you're right. I wanted a similar point to come across but as a question rather then a statement and managed to confuse myself in the process with regards to your answer. Macroeconomic analysis of Cell vs something else was sort of what i was getting at, a tough question.

Looking at it that way suddenly the potential value of Cell narrows significantly.

 

That's actually an interesting option. Cell is meaty enough for future game processing requirements. It's just being used to bolster RSX at the moment. Couple it with a good GPU and you'll actually have a capable platform that's BC. Sure, it won't be cutting edge performance, but as a viable economy option next-gen, it looks good to me. Developers will already have the tools and code and the experience, unlike trying to program new architectures. The GPU will work the same as all the others. It'd be a relatively straightforward system. The other alternative would be a straight multicore x86, which would be simpler to program but with lesser performance, and probably difficult hardware cost management (read Intel charging a huge markup!).

Much as I'd like a Cell2 64 SPU monster, a smaller improved Cell makes considerable sense in the cost-conscious future.