CELL: A nagging question

[Deepak]

A question is doing rounds in my mind. Its about this NV GPU, infact "GPU" itself. I remember reading articles about how powerful CELL will be, the focus was on CELL, it was supposed to drive PS3 to new heights. There was no metion of any GPU during those times. Now as CELL has been revealed, NV GPU has been mentioned, it looks like it would be the NV GPU that would make real difference and not CELL. CELL doesn't seem that significant/awesome anymore.

My question is...in Sony's initial scheme of things was CELL supposed to drive PS3 including GPU work, means there was no seperate GPU initially? And thats why they invested billions in CELL, but unfortunately maybe CELL didn't perform as well as they expected so they had to bring in NV? Maybe thats why Kutaragi was position was "downgraded"?


And regarding this alleged Toshiba GPU, is there any article (prior to NV annoucement) about it, I never read any article about it.

**********
All speculation on my part. Can anyone post other theories? Thanks!

 

[PC-Engine]

There are two theories. The first revolve around SONY having two plans right from the beginning, an initial plan (internally developed GPU) and a backup plan (externally developed GPU).

The second theory revolve around the existence of only a single plan (internally developed GPU). That plan didn't turn out as expected so an outside company was called in as a last ditch effort.

Both theories involved the internally developed Visualizer GPU based on the CELL architecture.

 

[Titanio]

AFAIK..

Cell was never going to be the sole graphics driver in PS3.

If you read IBM's docs on Cell, it makes it quite clear that the graphics workload they looked at when designing Cell was TnL, vertex operations. That's only one aspect of graphics. Cell in its current form isn't designed for pixel operations. As a vertex cruncher it should/could do very well, as a pixel cruncher it wouldn't - but it was never meant to. It would at least need a pixel cruncher paired with it - it was never designed to take on board all graphics operations. That is one of the biggest misconceptions in some quarters re. Cell.

This, btw, is where speculation has arisen that the CPU in PS3 will handle vertex operations, and that the NVidia GPU will be dedicated all or mostly to pixel operations.

Sony was always going to include a graphics chip in PS3 - the only question was from whom. People assumed it would come from Toshiba, but it seems now in hindsight that they were evaluating a number of competing options asides from Toshiba, and that they eventually picked NVidia over them.

 

[nAo]

:? :|

 

[Deepak]

:? :|

All speculation on my part

 

[Tsmit42]

I thought cell turned out exactly how it was meant to turn out:

1 SPE (APU) = 32 GFlops
1 PPE (PPU) = 256 GFlops

Well actually it turned out a bit better. There was always at GPU, cell based no doubt, but it was there.

 

[Titanio]

Well actually it turned out a bit better. There was always at GPU, cell based no doubt, but it was there.

There was a "Visualiser" diagram from one of the patents that featured Cell tech, but not for everything. It had SPEs but then it also had these unexplained "pixel engine" blocks - so presumably the SPEs would be doing the vertex work, and this extra hardware (albeit on the same chip), these "pixel engines" would be doing the pixel processing. So certainly, I don't think the presence of an externally designed GPU contradicts earlier expectations about Cell performance, more than it contradicts the notion that Toshiba/SCEI were the best people to provide the necessary hardware on the graphics side (or perhaps more specifically on the pixel pushing side)

 

[ShootMyMonkey]

Actually having a GPU of such design would be rather interesting in that it could mean having very flexible and programmable vertex processing before shipping of render-verts to the "pixel engines", which as far as the CELL patent was concerned, could have been any old rasterizer design as long as you had enough to perform. That could have opened up some avenues for mesh post-processing operations like subdivision or soft-body deformation and other fun stuff.

But I mean, unless the SPEs supported massive SMT so that you could hide all the memory access latencies, pixel processing is pretty much out of the question. I'm lost as to how having a GPU takes focus away from the CPU. A powerful GPU is simply a powerful GPU. No one with half a mind will ever use the GPU for AI, physics, game logic, etc. in a real commercial product. Yes, there's a whole lot of raw computational power there and it's interesting to try, but studies of general purpose computing on GPUs will forever remain academic exercises. In real performance-critical environments, things like academian interests and "proper" coding all get thrown out the window in the name of more speed.

 

[randycat99]

I remember reading articles about how powerful CELL will be, the focus was on CELL, it was supposed to drive PS3 to new heights. There was no metion of any GPU during those times.

Was there some legal obligation to fully disclose the entire system to us? Cell was the center of attention because it is a radical departure from conventional architecture. The potential "buzz" on something like that is priceless. Hence, it was the focal point of initial disclosure. They played the PR move exactly correct, imo. (I don't mean that facetiously or to imply that Cell is less than the specs suggest. It's just plainly a big talking point no matter how you slice it. Why not make full advantage of that?)

Now as CELL has been revealed, NV GPU has been mentioned, it looks like it would be the NV GPU that would make real difference and not CELL. CELL doesn't seem that significant/awesome anymore.

I believe that is mere perception at work, rather than fact. I don't see any new news that clearly cites that, yay or nay. The topic of Cell has certainly played it's day(s) in the sun, being discussed with great fervor. It's a completely natural process. Eventually, people will tire of discussing it. It doesn't mean it has become inconsequential to the design. When the tidbits of new info dry-up, the motivations to continue discussing Cell will dissipate, as well- a natural process.

Now there is more impetus to discuss the GPU, which is undoubtedly another crucial component in the whole system. It doesn't mean it has become the "central" part that "makes" the system. Simply, people have tired of talking about "Cell", and now the GPU is a new ball to kick around. That's really all it is, imo.

Maybe there will be a 3rd phase of interest and discussion when we get some info on exactly how the 2 work together. It's only a natural, logical progression. You can't really measure the success or failure of a single part simply because interest has waned after a big spurt in discussion (unless it is your goal to assert a meaning where there was no meaning in the first place- not an uncommon thinking process, at all).

My question is...in Sony's initial scheme of things was CELL supposed to drive PS3 including GPU work, means there was no seperate GPU initially? And thats why they invested billions in CELL, but unfortunately maybe CELL didn't perform as well as they expected so they had to bring in NV? Maybe thats why Kutaragi was position was "downgraded"?

Like I said, it's a cute theory. Maybe it's true. Most likely it is not. All we've got is a number of effects that can easily constitute a "story" just with the addition of some magical spin. I don't think it is a long walk to realize how this "deductive technique" gives us the mountains of jewel stories to fill gossip rag columns, no?

 

[BOOMEXPLODE]

It's alot like the PS2 announcements, where the "Emotion Engine" CPU was the focus of attention. That's probably just Sony promoting their own technology, just like Intel will promote its procesors and not the kind of memory they use. I think Mr. K. probably got into trouble for sinking so much of Sonys cash into Cell/PS3 development without regard for the bottom line (it was really alot of money) and so was taken out of the drivers seat. Developing the Cell based "Visualizer" would have been another huge investment and the obvious cheaper alternative would be to license a GPU design from a company like Nvidia. It was simply the most feasible course of action.