IBM unveils Cell roadmap

[Shifty Geezer]

GPUs at that time would be hard to go toe to toe with even with an Cell2 in your. Their "usable" capabilites, improved flexibility...

That depends entirely what direction Cell2 would go. But the idea isn't Cell as in PS3. It's a different Cell, with nVidia shader pipes (unified of course) for SPEs, or SPE's enhanced with nVidia shader tech.

I think the biggest problem isn't performance, but tools. At the end of the day, if the 'Visualizer' works as well as a standard GPU, why not use a standard GPU? And if it's better in some ways, how do you develop for it?

My guess is that if we're going to see a Cell with extra graphics abilities built in, it'll be in Sony's PSP2. Maybe something like a 1:4+2 PPE:SPE+VPE, clocked at 2 GHz. The biggest concern for a Cell VPE would be memory supply I think. The benefit would be the CPU and GPU on the same die with great flexibility and internal BW. Hmmm, maybe 4 MB of eDRAM on that chip too...

 

[dodo3]

Is it safe to say PlayStation Portable 2 will debut some day? Also, is it safe to say a smaller Cell chip will be implemented?

If Cell is implemented would there still be a need for a GPU? What kind of graphics/physics capabilities can we expect? Would it be easier to port next gen games because it is using a PPC proc? Could we see a higher resolution output or would it stay the same because it's portable?

Sorry a little off topic but I'd like some opinions.

 

[Arun]

Is it safe to say PlayStation Portable 2 will debut some day?

Yes.

Also, is it safe to say a smaller Cell chip will be implemented?

Maybe.

If Cell is implemented would there still be a need for a GPU?

Yes.

What kind of graphics/physics capabilities can we expect?

Much.

Would it be easier to port next gen games because it is using a PPC proc?

Not much.

Could we see a higher resolution output or would it stay the same because it's portable?

Maybe.

Sorry a little off topic but I'd like some opinions.

Yes. (and sorry, this reply style was just TOO tempting!)


Uttar

 

[Arun]

My guess is that if we're going to see a Cell with extra graphics abilities built in, it'll be in Sony's PSP2.

You can argue Sony is stupid, but I'd also like to argue they aren't completely retarded either. You would expect that IF NVIDIA got a contract in the PSP2, and the CELL is also part of it, the two architectures would be placed on the same chip. Think AMD Fusion: It's not because the two architectures are on the same die that their design is unified or vastlyh different. It does give you a cheap and very high bandwidth interconnect though, obviously, which has its own set of advantages.

How likely is it that NVIDIA will develop an architecture that is vastly different from their current one, when they haven't for RSX and yet got substantial (read: disproportionate, imo!) R&D money for it through NREs? And wrt using two CELLs intead of one CELL and one GPU, I see some people aren't quite learning their lesson...

You could definitely see a custom architecture where the VS and GS are done on the CELL and the PS is uber-optimized instead, but you'd hardly save any silicon compared to an unified architecture, so I'm not sure what's the point. It might happen for a variety of reasons, but technically it's not really a big advantage.


Uttar

 

[Shifty Geezer]

You could definitely see a custom architecture where the VS and GS are done on the CELL and the PS is uber-optimized instead, but you'd hardly save any silicon compared to an unified architecture, so I'm not sure what's the point. It might happen for a variety of reasons, but technically it's not really a big advantage.

That seems to be a point majorly overlooked. GPU's are actually very good at what they do. A different CPU architecture for Cell made sense to overcome limits of conventional designs in certain applications. There isn't a similar problem with GPU's that needs to be overcome with a new architecture, save perhaps the long tooted argument in favour of tile based deferred rendering, which has never really come to much.

At the moment the only reason I see for continuing to consider the possiblity of a Cell CPU+GPU hybrid are the comments from nVidia about they and Sony having the same vision. Which could well have been just a PR statement. But, if we look at the Cell Visualizer patents and the old idea of a CPU doing everything as Sony's long-term vision, it could be thought that nVidia want to get in on a hybrid processor, especially wth AMD's movements. Getting nVidia into Cell for when (if!) it makes the move into CE goods means nVidia are earning on every Cell device. If they don't manage that, Cell products will be free to pick and choose from other GPU tech.

Technically there might be little point to it, but I'm sure nVidia are keen on the idea!

 

[Arun]

As I said, the VS and GS are perfectly fine for CELL, and per mm², its performance is really acceptable compared to that of a modern GPU for that. If, and only if, you don't need heavy texturing for these operations, of course. Which remains to be seen.

I'd say CELL-like CPUs and GPUs are much more complimentary than fundamentally different. It's easy to see how 32 SPEs would be useful for heavily threaded workloads but not "massively parallel", while GPUs would remain better at data-level parallelism, where its native scheduling mechanism and its latency tolerance come into play. It's easy to see how CELL and a G100-level GPU would be extremely happy together on the same die.

I'm not sure how that fits in terms of economies of scale. Even on 45nm, you'd expect a 32 SPEs CELL to take more than 200mm², and you'd expect the GPU to take a similar die estate. 400mm² seems horribly unlikely to me, so unless 32nm is ready on time for PS4, which I doubt... And before someone brings up a hybrid project then, do you realize that would imply Sony would be giving up on CELL per-se? Err...


Uttar