PS3 to do real face morphing?

[SynapticSignal]

Demo's illustrate availabe power. And in that respect if Cell can render 100 ducks with full rigid body and fluid physics etc at 60fps, and a P4 can't

please give us facts, demostrate that a p4 can't


100 or 200 or 500 duks is not a "great power" trial if ou thinks that games as 99 night, kameo, and so on, features a number of real animated model with their physic and their AI in the range of
1000-4000 units at same time on screen

I think that a p4 can do the duck demo very easly
download the megaton demos 1.4, there's a demo with phsics, cloates and fluid simulation that run ato 60 fps on a celeron mendocino 500 Mhz
so, think what can do a P4 3.6 GHz

 

[one]

please give us facts, demostrate that a p4 can't


100 or 200 or 500 duks is not a "great power" trial if ou thinks that games as 99 night, kameo, and so on, features a number of real animated model with their physic and their AI in the range of
1000-4000 units at same time on screen

I think that a p4 can do the duck demo very easly
download the megaton demos 1.4, there's a demo with phsics, cloates and fluid simulation that run ato 60 fps on a celeron mendocino 500 Mhz
so, think what can do a P4 3.6 GHz

Where can I find megaton demos 1.4?

 

[Shifty Geezer]

please give us facts, demostrate that a p4 can't

You missed a bit...

Originally Posted by Shifty Geezer
Demo's illustrate availabe power. And in that respect if Cell can render 100 ducks with full rigid body and fluid physics etc at 60fps, and a P4 can't

I was explaining the purpose of demos which you seem to think worthless. Regards the rest of your points...

1) I don't for one second think Kameo or any other XB360 titles has mesh<>mesh physics interactions on that level. Games will use the simplest level required for realism.
2) Who says the Ducks demo was using Cell to 100% capacity? Perhaps it can handle 42 million ducks at once? The choice of ducks was a design decision in homage to the original PS2 duck demo.
3) I'm not a true A grade programmer, but I've written PC code using physics engines and they are far, far, far from that level of performance. I'm not sure a top-end PC processor could cope with that many cubes on a water surface at 60fps, let alone as well as rendering the graphics (no GPU) and performing the image processing needed for the EyeToy interactivity+fluid physics there. The Aegia PPU people writing the Novodex kit themselves have pointed out how much better Cell is at physics than a PC CPU. Unless the world is dillusional and all along a 500 MHz Celeron can handle complex geometry physics in realtime, I don't believe what you may have seen in the Megaton demo (without a link I can't reference it myself) was working in the same way as the LOD demo.
4) The duck demo disproves your original argument anyway

don't misunferstand me, I believe that cell can do
1) morphing from a cam
2) sound processing
3) phisycs simulations
4) rendering without a gpu
5) hundreds of ducks
6) AI processing

but in a demo, Cell have to process only one at time of this tasks

The LOD demo handled 3, 4 and 5 simulataneously, which wasn't only one at a time.

And finally
5) What are Sony and Toshiba supposed to do to advertise their product if you feel they shouldn't show demos? After all that's your point that demos are misleading and only marketting, but how else can you showcase your machine's performance except through visuals ways? Everyone's showed demos and I'm glad they do, as looking at nothing other than processor layouts and trying to guess what it can be applied to isn't going to win many positive opinions in the masses where you're trying to garner attention.

 

[SynapticSignal]

You missed a bit...

I was explaining the purpose of demos which you seem to think worthless. Regards the rest of your points...

1) I don't for one second think Kameo or any other XB360 titles has mesh<>mesh physics interactions on that level. Games will use the simplest level required for realism.

excuse me but...
what can change what you believe or what you don't believe?
what I think is that you can't know what level of physics will use games
in 99 nights and kameo the animated doll models (not static simple rigid ducks), fight one against each others other with full respect of physics, and against the player
in a demo of 99 night was showed as the physic works on a 1-2 thousand models (the player caused an explosion in the centre of the models)

sorry,I found the duck demo very MEH in comparison

2) Who says the Ducks demo was using Cell to 100% capacity? Perhaps it can handle 42 million ducks at once? The choice of ducks was a design decision in homage to the original PS2 duck demo.

because a demo is realized aiming to show or to demonstrate tha capability of a platform
is stuping believing that they've made a demo to show "the power of cell" so simple if the architetture is not bounded by this and cell can do much more

if find the ducks demo truly pathetic......

3) I'm not a true A grade programmer, but I've written PC code using physics engines and they are far, far, far from that level of performance. I'm not sure a top-end PC processor could cope with that many cubes on a water surface at 60fps, let alone as well as rendering the graphics (no GPU) and performing the image processing needed for the EyeToy interactivity+fluid physics there.

and you call "rendering" the ducks and the water on screen?
maybe 5-6 years ago, not in Q3 2005

Cell was suppposed to do more, processing vertex data to fetch rsx, the a simple "ghouraud ducks" that can be made by a ps1
came on...

and I'm not sure that the ghouraud shading was provided by cell, but I believe you

The Aegia PPU people writing the Novodex kit themselves have pointed out how much better Cell is at physics than a PC CPU. Unless the world is dillusional and all along a 500 MHz Celeron can handle complex geometry physics in realtime, I don't believe what you may have seen in the Megaton demo (without a link I can't reference it myself) was working in the same way as the LOD demo.

maybe you don't have undestood this simple fact:
I'm not talking of cell
I'm talking of pathetic ducks demo

I don't know if cell, using all the spe, the ppe etc, can reach an ageia performance
but back to real world:
It' can't do this while process vertex data, sound, AI, extra game computing, animation of the models, all via spe, simply can't

4) The duck demo disproves your original argument anyway

I don't think so
I'm waiting for some facts, please

5) What are Sony and Toshiba supposed to do to advertise their product if you feel they shouldn't show demos? After all that's your point that demos are misleading and only marketting, but how else can you showcase your machine's performance except through visuals ways? Everyone's showed demos and I'm glad they do, as looking at nothing other than processor layouts and trying to guess what it can be applied to isn't going to win many positive opinions in the masses where you're trying to garner attention.

because the have made demo useless

You have to focus on the important things, ok?

in the real world the things go different from what they've showed

I want to know how cell perform when It have to process game data, ai, physycs, animation, vertex computing, sound processing
I want to see a "playable demo" of a real world application
I want to see Cell that fetch rsx, via FlexIO, while the spe are busy for other tasks
I want to see the efficiency in real world

If they are not interested to show the power of ps3, but only cell (mah) why don't compile a version of "prime bench" (or other multi platform base bench) to linux/cell system?

why you don't say that the system that power the demos is a dual-cell @2.4GHz machine?

Are you biased or what?

 

[BlueTsunami]

excuse me but...
what can change what you believe or what you don't believe?
what I think is that you can't know what level of physics will use games
in 99 nights and kameo the animated doll models (not static simple rigid ducks), fight one against each others other with full respect of physics, and against the player
in a demo of 99 night was showed as the physic works on a 1-2 thousand models (the player caused an explosion in the centre of the models)

sorry,I found the duck demo very MEH in comparison


because a demo is realized aiming to show or to demonstrate tha capability of a platform
is stuping believing that they've made a demo to show "the power of cell" so simple if the architetture is not bounded by this and cell can do much more

if find the ducks demo truly pathetic......



and you call "rendering" the ducks and the water on screen?
maybe 5-6 years ago, not in Q3 2005

Cell was suppposed to do more, processing vertex data to fetch rsx, the a simple "ghouraud ducks" that can be made by a ps1
came on...

and I'm not sure that the ghouraud shading was provided by cell, but I believe you





maybe you don't have undestood this simple fact:
I'm not talking of cell
I'm talking of pathetic ducks demo

I don't know if cell, using all the spe, the ppe etc, can reach an ageia performance
but back to real world:
It' can't do this while process vertex data, sound, AI, extra game computing, animation of the models, all via spe, simply can't



I don't think so
I'm waiting for some facts, please



because the have made demo useless

You have to focus on the important things, ok?

in the real world the things go different from what they've showed

I want to know how cell perform when It have to process game data, ai, physycs, animation, vertex computing, sound processing
I want to see a "playable demo" of a real world application
I want to see Cell that fetch rsx, via FlexIO, while the spe are busy for other tasks
I want to see the efficiency in real world

If they are not interested to show the power of ps3, but only cell (mah) why don't compile a version of "prime bench" (or other multi platform base bench) to linux/cell system?

why you don't say that the system that power the demos is a dual-cell @2.4GHz machine?

Are you biased or what?

I think your waiting for an actual game. I would ask for you to wait for Spring 2006 to Fall 2006. You'll get your wish. For now...that was a tech demo.

 

[SynapticSignal]

I think your waiting for an actual game. I would ask for you to wait for Spring 2006 to Fall 2006. You'll get your wish. For now...that was a tech demo.

Lol, no

I'm asking a simple level demo, from any of the games in development
the ps3 will came out in spring 2006, so I think that the many of the launch games are near as complete

so I ask to see a little bit of real world application of cell

I say this because I know that apple discard cell in favor to x86 because of the lack of performances of cell in the real world apps

in the games we will see if a 512 KB cached, single-PPE in order processor can do well or not, do you agree?

 

[BlueTsunami]

Lol, no

I'm asking a simple level demo, from any of the games in development
the ps3 will came out in spring 2006, so I think that the many of the launch games are near as complete

so I ask to see a little bit of real world application of cell

I say this because I know that apple discard cell in favor to x86 because of the lack of performances of cell in the real world apps

in the games we will see if a 512 KB cached, single-PPE in order processor can do well or not, do you agree?

Thats what everyone is asking, but in all honesty. Your not going to get it. Sony is factoring marketing into their release of CELL in relation to PS3 information. Shifty has explained to you the importance of the Duck Demo but Demos are NOT meant to max out capabilites. Its meant to showcase the strength of a certain piece of technology.

Also, comparing it to 99 Knights is bad IMO, simply because your comparing a tech demo to a completed game. I would think you can tell the difference, but what your asking for is a game being compared to a game. At this point in time it won't be happening, possibly next E3 or some other convention in the near future. Until then, you've got a techdemo that shows physics, water simulation and alot of ducks. Take it as you will.

EDIT: You can also tell it was a "wink wink nudge nudge" situation in that they where showcasing the advancements over PS2 with how many ducks can be created and how they can all intereact with eachother, while being placed on water and a boat.

 

[one]

I'm asking a simple level demo, from any of the games in development
the ps3 will came out in spring 2006, so I think that the many of the launch games are near as complete

so I ask to see a little bit of real world application of cell

I say this because I know that apple discard cell in favor to x86 because of the lack of performances of cell in the real world apps

in the games we will see if a 512 KB cached, single-PPE in order processor can do well or not, do you agree?

I wonder why there's no technical demo of Xbox 360 CPU. Xenos is very interesting GPU, but we can't know how the CPU works in synthesized demos / games in which the both CPU and GPU work.
Do you happen to have any benchmark? Also where can I find megaton 1.4?

 

[Nesh]

please give us facts, demostrate that a p4 can't


100 or 200 or 500 duks is not a "great power" trial if ou thinks that games as 99 night, kameo, and so on, features a number of real animated model with their physic and their AI in the range of
1000-4000 units at same time on screen

I think that a p4 can do the duck demo very easly
download the megaton demos 1.4, there's a demo with phsics, cloates and fluid simulation that run ato 60 fps on a celeron mendocino 500 Mhz
so, think what can do a P4 3.6 GHz

Dont know much about it.But were the cell tech demos using only the cell processor?
If yes can a P4 alone without a GPU handle the megaton tech demo in software only?

Just asking to know.Not trying to argue.there

 

[SynapticSignal]

for All who are asking for the Megaton 1.4 demos (are a lot of physics, cloat simulation, fluid simulation PLAYABLE demos)

the website is
http://www.meqon.com/

but seems that ageia take the cake

the main directory is 5.5 MB, I've compressed to 1.35 MB
the demos was not copyrighted, so I can upload somewhere
hummmm where?

 

[Shifty Geezer]

excuse me but...
what can change what you believe or what you don't believe?
what I think is that you can't know what level of physics will use games
in 99 nights and kameo the animated doll models (not static simple rigid ducks), fight one against each others other with full respect of physics, and against the player
in a demo of 99 night was showed as the physic works on a 1-2 thousand models (the player caused an explosion in the centre of the models)

I can't know, but I can guess from knowledge and personal experience.2 colliding ragdolls each made up of thirteen bounding cubes is quick and easier to solve in a physics engine then 2 500 polygon (guess) arbitary meshes. Arbitary meshes are very consuming and in a complex pileup your talking huge numbers of collisions to test for.

Now I've already admitted that my experience is fairly limited and maybe you are better educated in this respect? If so it'd be good for you to post your credentials in coding with physics simulation and why you think collision solving with arbitary meshes is so much easier than ragdolls?

because a demo is realized aiming to show or to demonstrate tha capability of a platform
is stuping believing that they've made a demo to show "the power of cell" so simple if the architetture is not bounded by this and cell can do much more

Not at all. Demos are there as eyecandy too. It's a matter of showmanship. If they could have added 1 trillion ducks, you wouldn't have been able to see anything. There's the duck to tie in with the PS2's demo. Then there's more ducks. Then there's lots more ducks. It was a good showmanship piece at which the audience laughed and I believe were impressed. I guess they too think 50+ arbitary mesh collisions on a fluid surface is impressive.

and you call "rendering" the ducks and the water on screen?
maybe 5-6 years ago, not in Q3 2005

Well first of all I don't know what you were playing in 1999/2000, but if you think any old CPU can software render like that...maybe you haven't even watched the demo?

But more importantly...

Cell was suppposed to do more, processing vertex data to fetch rsx, the a simple "ghouraud ducks" that can be made by a ps1
came on...

...I now appreciate where you're coming from. You're original statement that demos aren't a fair way to showcase technology has changed into an argument that Cell was supposed to be a super amazing hypercomputer and it's fallen far short of your expectations. Which isn't the point of this thread. This crazy statement without any explanation of your reasoning...

It' can't do this while process vertex data, sound, AI, extra game computing, animation of the models, all via spe, simply can't

...also doesn't belong in this thread.

4) The duck demo disproves your original argument anyway

I don't think so
I'm waiting for some facts, please

you said that PS3 can't do rendering, physics and 100's of ducks at the same time, but the duck demo shows this. It's rendering 100's of ducks with full physics. Ergo your assertion that it can only handle one at a time is disproven.

because the have made demo useless
You have to focus on the important things, ok?

in the real world the things go different from what they've showed

I want to know how cell perform when It have to process game data, ai, physycs, animation, vertex computing, sound processing
I want to see a "playable demo" of a real world application
I want to see Cell that fetch rsx, via FlexIO, while the spe are busy for other tasks
I want to see the efficiency in real world

Well for starters how are they going to demonstrate RSX working when RSX isn't available yet? They wanted to showcase what they're new CPU is capable of doing. For real-world efficiencies we've got a few IBM papers and Toshiba realworld demos and we've got some working game demos running in realtime. Not a lot of info, but some nonetheless. And also, the whole point of this thread was a TOSHIBA demo of Cell, which you were saying was worthless as indicative of what the CPU can do, despite the fact the demo was exactly what the Cell was doing in realtime in a real-world software application. Why is a TOSHIBA demo of Cell worthless if it hasn't got FlexIO running and 7.1 audio encoding and AI? Why is a Toshiba demo worthless of what Cell is capable for Toshiba's plans for the chip just because it isn't a game?

If they are not interested to show the power of ps3, but only cell (mah) why don't compile a version of "prime bench" (or other multi platform base bench) to linux/cell system?

Yes, Toshiba are only interested in showing the power of Cell because they aren't selling PS3s. And do you REALLY think they should go to a consumer electronics expo with a dull as dirt collection of numerical Prime Bench stats? How much attention will that get? And does Prime Bench properly benchmark a CPU's capabilities in the areas Cell was primarily designed, like stream processing FP data?

Are you biased or what?

Yes. I'm biased against people starting a sensible discussion on one point only really with a different agenda. Your problem isn't the showcasing of demo's being unindicative of performance, but that you don't believe Cell offers any real tangible benefits over other solutions. The fact you believe a Celeron 500 can handle the LOD demo's physics shows you don't appreciate the calculations involved and are therefore in no real position to judge how much these demo's really showcase CPU performance.

When the demos first came out there wa the old bantering in this forum. There were those who questioned the validity of some of the demos shown (though I won't name those particular demos! ). I don't remember anyone else saying 'mah, LOD demo was rubbish. Any processor can do that.' I guess that makes us ALL biased

Anyway, if you want to talk about the E3 demos I suggest you search the forum and revist the old threads. Read those and then if you feel you've something to contribute go ahead.

 

[SynapticSignal]

Dont know much about it.But were the cell tech demos using only the cell processor?
If yes can a P4 alone without a GPU handle the megaton tech demo in software only?

Just asking to know.Not trying to argue.there

for what I Know, at E3 time there was no ps3 hardware ready, the only cell based machine was a 'server' with two cell @ 2.4 Ghz, and this is the machine behing the demos
I think that 1 cell do rendering, the other manage the physics (this was the original ps3 project, before sony kick off the toshiba gpu to take in the nvidia gpu)

if you search, you can find some images of the dual cell machine (seems a big big metallic cube)

 

[one]

for what I Know, at E3 time there was no ps3 hardware ready, the only cell based machine was a 'server' with two cell @ 2.4 Ghz, and this is the machine behing the demos
I think that 1 cell do rendering, the other manage the physics (this was the original ps3 project, before sony kick off the toshiba gpu to take in the nvidia gpu)

if you search, you can find some images of the dual cell machine (seems a big big metallic cube)

Eh?
The first PS3 devkit which was released in January has 1 Cell @ 2.4Ghz + GeForce.
The next version released in Spring 2005 too.
http://ps3.ign.com/articles/635/635630p1.html

 

[mckmas8808]

for what I Know, at E3 time there was no ps3 hardware ready, the only cell based machine was a 'server' with two cell @ 2.4 Ghz, and this is the machine behing the demos
I think that 1 cell do rendering, the other manage the physics (this was the original ps3 project, before sony kick off the toshiba gpu to take in the nvidia gpu)

if you search, you can find some images of the dual cell machine (seems a big big metallic cube)

Yeah SynapticSignal you really messed up there. You need to read the forum a bit more it seems.

 

[SynapticSignal]

I can't know, but I can guess from knowledge and personal experience.2 colliding ragdolls each made up of thirteen bounding cubes is quick and easier to solve in a physics engine then 2 500 polygon (guess) arbitary meshes. Arbitary meshes are very consuming and in a complex pileup your talking huge numbers of collisions to test for.

Now I've already admitted that my experience is fairly limited and maybe you are better educated in this respect? If so it'd be good for you to post your credentials in coding with physics simulation and why you think collision solving with arbitary meshes is so much easier than ragdolls?

what do you not understand of the fact that I'm talking of 2000-4000 (two-four thousand, impress in you mind this numbers) ragdolls
try to do some math
I'm a coder too by 22 years ago, in different field, do you need my curriculum to understand what I'm saying?

a) 2000 to 4000 animated, each with AI models on a screen that fights and reacts to physics forces
b) 100 simple, gourauded, plain, static, rigid ducks on a screen

and I remember that there're two cells in the machine used for the demos, one does rendering, the other physic simulation

Not at all. Demos are there as eyecandy too. It's a matter of showmanship.

easy time
there's no eyecandy
the render is pathetic, the iq is pathetic, where's the eyecandy of the ducks?

I hope that you're just jokin'

If they could have added 1 trillion ducks, you wouldn't have been able to see anything.

then, instead to make 1 trillion of trillion of fantastillion of duck in a simple demo, why don't make 4000 animated ducks in a lake that react to ambience and one to each other?
I think to know the answer, you don't?

There's the duck to tie in with the PS2's demo. Then there's more ducks. Then there's lots more ducks.

here I've thought "PlayStation 2.5 here"

Well first of all I don't know what you were playing in 1999/2000,

a PS2?

but if you think any old CPU can software render like that...maybe you haven't even watched the demo?

you seems confused

I'm talking of the ghouraud rendering, and I've say "PS 1", but you undestand "software render"

read slowly please

But more importantly...

...I now appreciate where you're coming from. You're original statement that demos aren't a fair way to showcase technology has changed into an argument that Cell was supposed to be a super amazing hypercomputer and it's fallen far short of your expectations. Which isn't the point of this thread. This crazy statement without any explanation of your reasoning...

...also doesn't belong in this thread.

the thread topic is "ps3 to do real face morphing", but you're talking of physics and duck demos,
who is off topic, friend?
are you really confused eh?

you said that PS3 can't do rendering, physics and 100's of ducks at the same time, but the duck demo shows this.

again, you are getting confusion on confusion
you say that RSX is not available yet, but you say that the duck demo shows that PS3 can do this
PS3 without rsx?
hey, are you there, friends?
you're saying a lot of senseless things, the PS3 demo of E3 runs on a dual-cell computer

Well for starters how are they going to demonstrate RSX working when RSX isn't available yet? They wanted to showcase what they're new CPU is capable of doing. For real-world efficiencies we've got a few IBM papers and Toshiba realworld demos and we've got some working game demos running in realtime. Not a lot of info, but some nonetheless. And also, the whole point of this thread was a TOSHIBA demo of Cell, which you were saying was worthless as indicative of what the CPU can do, despite the fact the demo was exactly what the Cell was doing in realtime in a real-world software application. Why is a TOSHIBA demo of Cell worthless if it hasn't got FlexIO running and 7.1 audio encoding and AI? Why is a Toshiba demo worthless of what Cell is capable for Toshiba's plans for the chip just because it isn't a game?
Yes, Toshiba are only interested in showing the power of Cell because they aren't selling PS3s. And do you REALLY think they should go to a consumer electronics expo with a dull as dirt collection of numerical Prime Bench stats? How much attention will that get? And does Prime Bench properly benchmark a CPU's capabilities in the areas Cell was primarily designed, like stream processing FP data?

so first you say that the demo shows what a ps3 can do, and now you say that "it is only a toshiba Cell demo" (another thing false, it is a sony demo)

Yes. I'm biased against people starting a sensible discussion on one point only really with a different agenda. Your problem isn't the showcasing of demo's being unindicative of performance, but that you don't believe Cell offers any real tangible benefits over other solutions. The fact you believe a Celeron 500 can handle the LOD demo's physics shows you don't appreciate the calculations involved and are therefore in no real position to judge how much these demo's really showcase CPU performance.

my thoughts are clear, I don't need transaltion, thanks
I say that the duck demo is pathetic

and I've a LOT of doubts on Cell efficiency (Apple agrees with me, seems)

after all you have to remeber few heavy points (look at the datasheets)

1) Cell is provided with only 512 KB of cache
2) spe can't reach directly the main memory (they works well when a spe fetch the next one, as in multimedia processing, but very bad in the others fields)
3) Cell has an hit of 90% processing integers
4) Cell has an hit of 90% processing double precision numbers
5) Cell is an in-order cpu
6) Cell has only 1 PPE for general purpose computing

this are facts, if know a bit of the cpu arch. you can ready undestand why what I can see is a cpu that can do great thing on a videocam, on a HD-DVD / BR player, on a phone of last gen, and so on
but a very bad and inefficient solution for general computing (PC, servers, ask to apple if you don't believe), and a bad cpu for a console
the only, the ONLY thing that can save Cell on a console is using the spe to process dinamically vertex data on textures and fetch the gpu with a constant stream

but here comes the problem that I was talking
Cell have to do sound processing, AI, physics, game code processing, vertex
how can the poor spe do all this if they can't reach directly the sys mem??
you think that a little 512 KB of cache can be enought to store all the data requested to the spe?
512KB that are used by PPe too

I don't think that a thing as is this can do wonder, IMO It can do very bad
don't forget that apple have discarded the Cell Project just because of the dissapointing for performances........

 

[SynapticSignal]

Eh?
The first PS3 devkit which was released in January has 1 Cell @ 2.4Ghz + GeForce.
The next version released in Spring 2005 too.
http://ps3.ign.com/articles/635/635630p1.html

Her
I'm not talking of the dev-kit
I'm talking of a dual-cell machine used to show the demos

 

[mckmas8808]

Her
I'm not talking of the dev-kit
I'm talking of a dual-cell machine used to show the demos

What dual-cell machine used to show demos? What demo was that? I think you are thinking about the terrain demo; which I actually read that somebody told me IGN got wrong. They supposely thought the demostrator was talking about two CELL chips when they weren't.

 

[SynapticSignal]

What dual-cell machine used to show demos? What demo was that? I think you are thinking about the terrain demo; which I actually read that somebody told me IGN got wrong. They supposely thought the demostrator was talking about two CELL chips when they weren't.

the machine under the ps3 demo is a mercury server
and when they talk of terrain demo (this is one of the demos that ran on the mercury machine at E3) they have stated that "one cell do rendering, the other do terrain simulation"

PS. the mercury dual-cell will came to the market soon, so we can finally do some advanced real testing

 

[BlueTsunami]

the machine under the ps3 demo is a mercury server
and when they talk of terrain demo (this is one of the demos that ran on the mercury machine at E3) they have stated that "one cell do rendering, the other do terrain simulation"

PS. the mercury dual-cell will came to the market soon, so we can finally do some advanced real testing

I believe the Terrain Simulation is probably correct. E3 though has always been a single CELL with an Nvidia GPU.

 

[Edge]

> "and I've a LOT of doubts on Cell efficiency (Apple agrees with me, seems)"

That's totally false. There was a number of concerns, with the main one being, porting code to a new architecture. The Intel porting tools have already been five years in the making, and also the size of the Intel CPU market versus CELL was a major consideration. You going to get a lot more options from x86 than CELL. More processor designs to choose from for different segments (Desktop, notebook, small form factor, etc.)

Efficiency was only mentioned if running all code off of the PPC, and why would anyone use a processor ignoring 90 percent of it's power.

Porting over Apple's software to take advantage of CELL strengths would have taken years.

CELL is certainly not inefficient, but an incredibly powerful monster, when you write good code for the SPE's. A number of benchmarks already show it running circles around traditional architectures when running multi-media based code which is important to the PS3's success.

> "3) Cell has an hit of 90% processing integers"

Totally misleading, as all integer code can be vectorized at compile time. The only issue is if you need an integer answer, there is extra work involved as you have to convert back.

> "4) Cell has an hit of 90% processing double precision numbers"

Again totally misleading. Why is double precision so important to a game console? It's not!!!

> "1) Cell is provided with only 512 KB of cache"

And almost 2 Megabytes of localized memory!!! Did you forget that? Read my signature line, and you will understand why CELL is such an impressive architecture. SEVEN seperate CPU's with localized memory.