Cell on NVIDIA Graphics Cards?

[rwolf]

There has been talk about cell having poor performance. Much lower than expected. I think it was originally envisioned to do the graphics processing as well and then at the last moment they had to cut a deal with Nvidia because the performance wasn't cutting it.

 

[Tim Murray]

Let's alter the question slightly. If you were Nvidia and you were looking to have a Physics processor on board, where would you look? There are a couple of things here:

At the vector processor I already make?

 

[rwolf]

At the vector processor I already make?

Exactly, in fact you would probably not want to dedicate transistors unless you could resuse them for other purposes hense the unified pipeline.

 

[nutball]

- having the physics processor close to the gpu seems to me pretty important. The Ageia chip has trouble partly because it can't very efficiently interact with the graphics chip.

One of the criticisms levelled at the NVIDIA/ATI strategy of using the GPU for physics is that it's mostly about superficial "eye-candy" physics, rather than gameplay-affecting physics. This is mostly because the engine driving the gameplay (ie. the CPU) has real problems accessing the current state of the physics system on the GPU.

It's not clear to me how moving the physics system onto a Cell on the graphics card will solve this problem. Giving a Cell on the graphics card the responsibility for the gameplay too might solve this issue -- but in that case what the hell is the CPU going to have left to do?

It seems to me that if you want gameplay-affecting physics it's important to have the PPU closely bound to the CPU not the GPU.

 

[_xxx_]

Well, I take your point, but would argue that it doesn't always apply. Intel took the P4 from 180nm to 65nm without integrating any significant extra functionality onto the die.

In case of the PC, it'd rather be the chipset which gets more and more features.

 

[_xxx_]

There has been talk about cell having poor performance. Much lower than expected. I think it was originally envisioned to do the graphics processing as well and then at the last moment they had to cut a deal with Nvidia because the performance wasn't cutting it.

They wanted to use cell only without an extra GPU and that's where it didn't live up to the expectations. As "just" a CPU, it rocks.

 

[nAo]

They wanted to use cell only without an extra GPU and that's where it didn't live up to the expectations. As "just" a CPU, it rocks.

I wonder if someday this urban legend will die..

 

[_xxx_]

I wonder if someday this urban legend will die..

Huh? AFAIK Sony people mentioned that numerous times in interviews and such?

 

[Robert.L]

Huh? AFAIK Sony people mentioned that numerous times in interviews and such?

MS and ATI don't count as Sony people

 

[Skrying]

MS and ATI don't count as Sony people

What? I'm fairly certain that it was indeed what Sony wanted to do.

Really, what's so hard to believe about it?

 

[Robert.L]

What? I'm fairly certain that it was indeed what Sony wanted to do.

Really, what's so hard to believe about it?

Because it wouldn't even take a second for an engineer to realize that another cell just wouldn't cut it as a gpu.

I’m sorry i came off a tad bit rude in my post above but this whole myth of another cell in ps3 acting as a gpu is just myth like nAo said.

 

[_xxx_]

Not "another" Cell, they wanted to use just one of them. And we're talking about 2001/2002 timeframe.

 

[nAo]

Not "another" Cell, they wanted to use just one of them.

emh..no, there was no project to use one additional CELL or just one of them as GPU.

 

[Titanio]

Well, there was the idea of that..though it may not have gone very far.

One of our ideas was to equip two Cell chips and to use one as a GPU, but we concluded that there were differences between the Cell to be used as a computer chip and as a shader, since a shader should be graphics-specific. The Cell has an architecture where it can do anything, although its SPE can be used to handle things such as displacement mapping.

Cell is a good renderer, but the types of rendering it may be better than a 'traditional' GPU at aren't exactly established as the norm in videogames or realtime rendering (yet, anyway). Although I think it will be interesting to watch what happens in PS4, in terms of the relationship between the CPU and GPU.

 

[KimB]

Cell would be a couple of orders of magnitude slower at 3D graphics than a modern GPU.

 

[Titanio]

Cell would be a couple of orders of magnitude slower at 3D graphics than a modern GPU.

There are certain types of rendering it appears to be very good at. Volume rendering, ray-casting etc. I expect it would be quite slower with the rendering approach of a typical modern GPU, though, yes - it's certainly not great at all for general texturing, probably not as good at rasterisation or fragment shading as it as geometry processing etc. If Cell were to be used today as a GPU, it would need a lot of specialisation. I don't know where we'll be in 6 years though..

 

[KimB]

Both of which would look like crap compared to rasterization without much, much more processing power than Cell has available.

 

[Titanio]

Depends on how where you apply it But the context today still requires a GPU for everything else to look good with reasonable efficiency, so your point is taken.

Still, my (admittedly foggy) crystal ball does wonder how things will go from here if CPUs (like Cell at least) and GPUs grow closer.

 

[zed]

Cell would be a couple of orders of magnitude slower at 3D graphics than a modern GPU.

true with 7/8 cores but by next console generation how many cores will it have?
i expect 1024 cores would be enuf to do realtime raytracing (dont know if 1024 will be achievable by year 2013 though)
the merging of graphics/cpu is inevitable the only question remains when?

 

[KimB]

Except that GPU's will have evolved by a similar degree, and I don't think that process technologies are moving that fast any longer.