Details trickle out on CELL processor...

[Panajev2001a]

Still say a TFLOPS BE is imposible?

No, but I still say that PlayStation 3 will not pack a 1 TFLOPS CPU for a whole host of reasons .

 

[j^aws]

We still don't know how many Cells are there in the Workstation to give us an indication of the capabilities per Cell.

If current 1st gen WS are 2TFlops, and one Cell is 294.4 GFlops, then around ~ 7 Cells/PEs per Workstation?...that's a strange odd number, perhaps 8 Cells?

 

[Qroach]

Yeah...PRman has pretty much nailed with Pixar, especially with it's raytracing/GI module...

As far as I can recall renderman has "support" this for years yet pixar rarely used it. In fact I seem to recall pixar saying something to this affect somewhere...

 

[j^aws]

Yeah...PRman has pretty much nailed with Pixar, especially with it's raytracing/GI module...

As far as I can recall renderman has "support" this for years yet pixar rarely used it. In fact I seem to recall pixar saying something to this affect somewhere...

Maybe the algorithms were resource hungry/inefficient back then...?

 

[Jov]

I'm curious where did this 15 TFLOPs workstation come from?

And why would Sony be sending 15 TFLOP workstations for a console thats only 1/15th that at most? (heh I sure hope its not so we can have really nice CGI cutscenes)

The Register link Jaw's points to sounds definately more believable.

I have to say that rack looks nice and maybe the multi-threading is their idea of hyperthreading? Just throwing out a wild guess with that one.

At 90nm, was it 16 TFlops for a rack mountable machine and 2 TFlops for the Workstation?

 

[j^aws]

I'm curious where did this 15 TFLOPs workstation come from?

And why would Sony be sending 15 TFLOP workstations for a console thats only 1/15th that at most? (heh I sure hope its not so we can have really nice CGI cutscenes)

The Register link Jaw's points to sounds definately more believable.

I have to say that rack looks nice and maybe the multi-threading is their idea of hyperthreading? Just throwing out a wild guess with that one.

At 90nm, was it 16 TFlops for a rack mountable machine and 2 TFlops for the Workstation?

STI press release, 16TFlops "will reach", implies 2nd gen. The Register mention "prototype" is 2TFlops, implies, they're the ones going to devs as 1st gen.

 

[Jov]

We still don't know how many Cells are there in the Workstation to give us an indication of the capabilities per Cell.

If current 1st gen WS are 2TFlops, and one Cell is 294.4 GFlops, then around ~ 7 Cells/PEs per Workstation?...that's a strange odd number, perhaps 8 Cells?

Don't we expect the PS3 to be roughly ~ 1/2 the capability of the Cell WS given it will be used as the dev platform?

 

[j^aws]

We still don't know how many Cells are there in the Workstation to give us an indication of the capabilities per Cell.

If current 1st gen WS are 2TFlops, and one Cell is 294.4 GFlops, then around ~ 7 Cells/PEs per Workstation?...that's a strange odd number, perhaps 8 Cells?

Don't we expect the PS3 to be roughly ~ 1/2 the capability of the Cell WS given it will be used as the dev platform?

Best for a resident PS2 dev to confirm but, IIRC, PS2 TOOL has the same power as PS2 but with more RAM and other i/O stuff...> PS3 = 2TFlops without a GPU?

 

[binary42]

Raytracing

Check out http://www.openrt.de/ .... They have been doing clustered real-time raytracing via clusters for quite some time now. I went to a presentation from the group that organized this project and they mentioned Sony had talked to them about raytracing... not many have beleived me since then but this new information makes perfect sense. Raytracing is overhaul if you plan on continuation of current level of graphical effects BUT there are many things that you can't do *easily* w/o a form of raytracing. I doubt complete raytring will be used but I believe that it will be a used in some form.

Here to hope and even if this is not 100% real even then the specs are most imressive. I think Sony has more credit this time around as they decided to not do an in house solution but go with IBM and Toshiba on this one. IBM does hype but when it comes to semiconductors they usually deliver what they say they will. I do hope they get good yeild rates to make this more affordable because I want one!

 

[binary42]

Jaws, I am not a pro Ps2 dev. but I do a lot of homebrew. The specs are basically available online for the T10000 (how many zeros again, I think its 10k?) dev kit. It is basically more RAM AFAIK.

 

[j^aws]

Jaws, I am not a pro Ps2 dev. but I do a lot of homebrew. The specs are basically available online for the T10000 (how many zeros again, I think its 10k?) dev kit. It is basically more RAM AFAIK.

Thanks...I've read that stuff on SaarCOR a while ago...It's definitely cool stuff and indeed it did remind me of Cell! ...<Crosses Fingers and Toes>

 

[Jov]

Don't we expect the PS3 to be roughly ~ 1/2 the capability of the Cell WS given it will be used as the dev platform?

Best for a resident PS2 dev to confirm but, IIRC, PS2 TOOL has the same power as PS2 but with more RAM and other i/O stuff...> PS3 = 2TFlops without a GPU?

Didn't KK stated 1000x the PS2 (not that I'm holding it against him), is it even remotely possible to think the PS3 might actually turn out to be that on raw specs alone, peak?

Cell WS with 2TFLops + G/VPU @ 90nm ~ 3 - 4 TFlops depending how it’s measured, thus 4+ TFlops @ 60nm.

Even if the WS meets the dream specs, the next question is will it fit in the PS3 given all the cost and space constraints?

 

[Deepak]

Wow, Cell Info. I browsed through some articles and each presented a different picture. Can anyone post the main points?

 

[PC-Engine]

Still say a TFLOPS BE is imposible?

Nobody ever said it was impossible. What people were saying was that a 1TFLOPS single CELL chip with 32MB of eDRAM, 1MB of SRAM, 4GHz operating frequency, less than 300mm^2, and aircooled in early 2006 for PS3 is impossible.

 

[V3]

Sorry, but this really doesn't make any sense. It's a completely arbitrary restriction, imho. Even a so called stream processor (by your defintion) like a
vertex shader engine could kill triangles by collapsing different vertices in degenerate primitives.

It was just a simple explanation. If those APUs turn out to be stream processor there are ways around that problem. I am just suprise if it is, since the patent didn't give any hint of it being one.

Are pixel shaders hw implementations on current GPUs stream processors?
I believe they are..but they can arbitrarily kill pixels!

Most of them are, but there are probably units there that aren't stream in nature I belive.

 

[archie4oz]

Yeah...PRman has pretty much nailed with Pixar, especially with it's raytracing/GI module...

Eh? The GI support in PRman is pretty recent (and kinda ghetto too I might add, although there's been significant improvements in 12.0)...

As far as I can recall renderman has "support" this for years yet pixar rarely used it. In fact I seem to recall pixar saying something to this affect somewhere...

Just a couple at most... Any raytracing/GI use before, they were using BMRT as a ray server for PRman...

 

[Mortimer]

Couple points I saw in those press releases:

The first computing application IBM plans for Cell is the Cell processor-based workstation it is developing with SCEI.

Looks like there is no workstation yet.
But on the other press release:
IBM, Sony Corporation (Sony) and Sony Computer Entertainment Inc. (SCEI) announced today that they have powered-on the first Cell* processor-based workstation.

And the performance:
The companies expect that a one rack Cell processor-based workstation will reach a performance of 16 teraflops or trillions of floating point calculations per second.

One WHOLE rack of 48 units high? 24 units? 1 unit?
Just how many cells does it take to get that performance?

There just isn't any real info on those press releases.
Wait for february.

 

[nAo]

One WHOLE rack of 48 units high? 24 units? 1 unit?
Just how many cells does it take to get that performance?

A first generation CELL core should be composed of one PU (a POWER CPU) and 8 APUs. An APU should peak at 8 floating point operations per second.
From that ISSCC slide we can project a working frequency for the CELL CPU of at least 4 Ghz.
If all these assumptions hold true a single first generation CELL core should peak at 256 Gigaflops/s. To make a 16 TeraFlop/s workstation they need 64 processors.

ciao,
Marco

 

[PC-Engine]

How does this CELL chip compare to that of 3DLabs P20?

 

[Mortimer]

One WHOLE rack of 48 units high? 24 units? 1 unit?
Just how many cells does it take to get that performance?

A first generation CELL core should be composed of one PU (a POWER CPU) and 8 APUs. An APU should peak at 8 floating point operations per second.
From that ISSCC slide we can project a working frequency for the CELL CPU of at least 4 Ghz.
If all these assumptions hold true a single first generation CELL core should peak at 256 Gigaflops/s. To make a 16 TeraFlop/s workstation they need 64 processors.

ciao,
Marco

Lots of assumptions when it seems like they themselves don't seem to know the performance yet and whether they have a working workstation or not.

But come february, we will be much wiser.
What I'd really like to know now would be the size of this first gen cell in 90nm with one PU and 8 APUs. From that size we could figure out what could be the PS3 version of it in 65nm.