Kutaragi talks CELL technology @ TGS2002

I'm running a GHZ TBird, GF2Pro and 512MB RAM(SDR). Drop your AGP aperture to 4MB in your BIOS, seems to be a problem with the game whenever you have it set higher(it was running like absolute crap when I had it set at 64MB, almost as choppy as the PS2 port).

 

Don't forget these numbers are an average, not a bare minimum.

73.7fps on UT2003?? It absolutely chugs on my Duron 1Ghz.. I have to turn the detail settings down to a bare minimum to get anywhere near 60fps.

My original point still stands and I believe holds true: there is no way a 1Ghz PC has a real-world (i.e. gaming) advantage of 2x-3x over the PS2, if anything the console can get better performance because of its fixed structure.

 

As the PS2 port? Que?

Anyway GTA3 runs really well on my PS2, but terrible on my friend's PS2 - I believe it depends on the condition of your DVD-ROM. He has a lot of problems with movies and his load times are longer, so I think that has something to do with it.

Plus, let's not forget than in the case of GTA3 it's mostly hampered by its streaming efforts - and you're running it ona PC with 512MB of RAM, compared to 38MB

 

I already stated that I did not intend to claim said pc (as a system) has 2x-3x the "real world" performance as PS2. I was referring to computational performance of both cpus, so "real world performnce" was definitly a misleading term.

 

You can't directly compare them though, because the PS2 is meant only for game playing - nothing else.

And in that regard, it is capable of much better visuals with what was available on PC at the time.

 

The answer to wether or not ps3 will come close to 1000xps2 is very simple, it can be rephrased into....

What is the purpose of the ps3 hw in relation to the playstation brand?

and the answer to that is pretty simple... if we go back to the xbox announcement and how sony overeacted, shocked by the fact a competitor had announced a more powerful product... we can begin to see this answer...

The purpose of the ps3 hw is to allow the playstation brand to have a comparable performance to a product that could come even 2yrs later, this is to allow for 1-2yrs lead, cause gphx hype crush if anything come before it, and if anything comes soon after it wont have anything above the playstation product... at least that should be the real intent behind anyone who is leading the console market, and seeing sonys reaction it is very clear this is what they intend...

Last time they freaked out, now they have the money, the time, and the engineers to achieve performance that wont be significantly eclipse by a product that comes 1-2yrs later, and secure their gphx front in the next console wars.

 

Yes, because we all know ... if you throw enough engineers at a problem such a piddly hurdle as physics moves right out of the way and lets you put a 1000$ bucks worth of silicon and a 1000 Watts worth of transistor switching into a usable device.

 

Correct me if I am wrong here, but here is my current understanding of CELL: A CELL chip will contain between 8 and 32 CPU cores (maybe only 8 ) on a single die, depending on the intended use. but from what have heard, the version of CELL that might be used in PlayStation3 (thus what the Emotion Engine 3 could have) is a 4-CELL chip - This is the thing that would have 500 million transistors, as Sony mentioned EE3 would have 500M some years ago. (around 1999)

So the EE3 for PS3 could have between 32 and 128 CPU cores, (four CELLs in one) depending on the complexity of the basic CELL. Even if EE3/PS3 had as few as 32 cores (four CELLs each with 8 cores) thats still a **** load of CPU power, if clocked at 1 Ghz.

 

About IBM's Power4 - each 1 Ghz Power4 chip has 2 64-bit PowerPC CPU cores conected by a 500 Mhz bus - this is the basic building block - IBM produces a 4-way Power4 SMP - so that's FOUR 2-CPU chips (8 CPU cores) with lots of eDRAM or caches (forgot if its eDRAM or caches) - The entire 4-way Power4 SMP is several inches across, IIRC - This is how I think PS3's Emotion Engine 3 will turn out -

EE3 might be something like 4 CELL chips on a single die - each CELL chip will have 8-32 CPU cores - This is the only way Sony is going to get MULTI TeraFLOP performance from PS3. (2-5 TFLOPs) The EE3 would probably be clocked at 1 Ghz, maybe as high as 1.5 Ghz. of course, that is all speculation, and it could all turn out differently, but something along those lines is the only way that would allow Sony to reach its stated goal of Playstation3 being 1000x the performance of PS2.

The GS3 would be the other part of the equasion. something like 128-256 pixel engines (GS has 16 pixel engines) with 32-64 MB of eDRAM, clocked at 500 Mhz (if EE3 is 1 Ghz) or 750 Mhz (if EE3 is 1.5 Ghz)

My ideal senareo would have EE3 clocked at 3 Ghz (thus clocked 10x higher than EE, like EE was 10x higher than PSX's CPU) and the GS3 would be clocked at 1.5 Ghz - then Sony might reach 1000x PS2 performance. but these speeds are highly unlikely. More likely, 1 Ghz for EE3 and 500 Mhz for GS3. I'd be impressed even if Sony gets 500x PS2's power in PS3.

 

There's a lot of ifs there.

 

I fail to see the logic here: why would Sony want to sell a supercomputer for $300?

And most of the guys here are pretty quick to drop the software problem. What's the worth of the best hardware if you can not program it?

 

SCEA doesn't want to sell a $300 supercomputer, just to have a supercomputer. Sony wants to have a machine with a few chips in it that can push the MOST amount of polygons at a massmarket price. It wants chips with enough computational power (EE3) and rendering power (GS3) for the most realistic images possible that cannot be matched by its competitors at the time of its release. that is what Sony has tried to do historically.

 

bump

 

Question: How much computational power does Ray tracing and or Photon tracing require?

Is the theoretical power of CELL enough for those calculations?

 

ok newbie question

What is the definition of ray-tracing and photon-tracing?

 

They are ways of modeling light by following the path of light particles (photons) or rays. Simply put, it gives a render a VERY realistic apearance even if the scene is just comprised of a single model of a cube or sphere. Doing it in realtime like at 60 fps takes A LOT of processing power.

http://graphics.lcs.mit.edu/~henrik/pictures/imgs/cboxgi.jpg

http://www-cse.stanford.edu/classes/sophomore-college/projects-97/ray-tracing/types.html

 

How long does a medium-complexity scene (such as you would see in a better or future videogame, that is) take to render offline on "today's computer", using traditional raytracing?

 

It depends on resolution and scene complexity. There are many visual demos that perform realtime ray tracing, with more or less success. One of the most prominent examples is:

ftp://ftp.scene.org/pub/parties/2000/mekkasymposium00/in64/h7-final.zip

Keep in mind, this is running purely on CPU - it does not require 3D accelerator card of any kind.

On my 1GHz machine, it's running smoothly at 320x240, choppy at 640x480 and even worse at 800x600

 

Kinda interesting trivia - Cell's roots began as an IBM sponcered raytracer. The idea being to break the task up into many individual threads and have each 'ray' be computed on a cut down processor core with it's own local embedded ram.

 

Trouble with raytracing is, that as you start doing refraction/reflection you rays suddenly loose all spatial locality and hence, individual processor has to be able to access the entire world dataset.

Cheers
Gubbi