ND Interview - PS3 to have bigger improvement between gens.

I\'d have to actually care about what you post for it to hurt my feelings. I just skim over your posts and dismiss them all as third world country ignorance.

:thumb:

Too bad for you, you do

[BizioEE]

I believe in the huge talent of ND, and Uncharted is simply the proof they know what they\'re talking about !
The Cell is a beast, and it\'s a must that the most talented developers will push more and more year by year this hardware, and ND is one of the most talented! Not surprised, just expecting!

I\'m ready to place a bet with anyone that I\'ll cry when the sequel of Uncharted will come out !

PS : finished Uncharted at normal, now playing at hard...60%...waiting to play at "ultra hard"...after I finish at hard, I presume...

[BizioEE]

Naughty Dog uses a Job Manager developed jointly by Naughty Dog’s ICE team and SCEE’s ATG group. This means that we can send any type of job to any SPE, and all of the scheduling of jobs is done through a priority system. This works well, since the overhead is minimal and we achieve good load-balancing between SPEs, something that would be hard to do by allocating a whole SPU to a single task.

Source: n4g.com

It appears that Cell is a perfect match for procedural texturing and ray tracing, as witnessed with these ProFX middleware technology benchmarks. ProFX is now in some form, apart of PS3 middleware.

The Bayou benchmark below was completed in 10 seconds by the 360, 5 seconds by the 8800GTS and 1.2 seconds by the PS3.

Here\'s an interview with Sebastien Deguy, president of Allegorithmic.


Quote:
Graphics, Games, and Multi-core
Jonathan Erickson
The challenges for developers may not be just technical

Today\'s guest is Sebastien Deguy, president and founder of Allegorithmic.


DDJ: Sebastien, the Allegorithmic web site mentions "procedural textures." What does that mean?

SD: Procedural textures are textures defined according to an algorithm. Instead of painting (by hand) the texture pixel by pixel, you define the way these pixels have to be lit to produce the texture you want. It\'s basically like writing a program, and with that program comes all the power of algorithms over frozen (bitmap) data -- compactness, data amplification, parameterization, and the like.

When the procedural texture is defined, you launch an engine for actually generating the bitmap textures. These are called realizations of the procedural description, and it\'s a lot like the difference between a -- possibly random -- process and the realization of that process.

In the case of Allegorithmic\'s procedural textures, you\'d define the procedures in a nice visual way using our product MaPZone, then use our other technology ProFX to rasterize the images, producing the final result to be displayed by a game engine or any renderer.

There are many advantages to using procedural textures instead of bitmap textures, one of them being that procedural texture files are typically 500-1000 smaller than bitmaps (making them ideal for online games).

DDJ: I\'m familiar with 3D textures. But what are 4D textures?

SD: 4D here stands for 3D + Time. 4D textures means evolving textures, evolving environments and games. On our web site we showcase a demo around that idea, where a bathroom goes from a clean state to a dirty state in real-time.

Procedural textures, because they are driven by parameters, can be driven by time. And by letting time go by, the aspect of the textures can be changed, going for example from a clean state to a deprecated one.

Now imagine that applied to a game, where the whole environment and characters are changing, evolving, according to time but also to your actions within that environment. Actual PCs and consoles can handle this, and I must say I can\'t wait to see games using this as a feature.

DDJ: If I recall from your web site, a platform you will shortly be supporting is Playstation 3. What benefits and challenges do multi-core processors bring to the scene?

SD: Procedural techniques are demanding in terms of power, so the more power you can get, the better it is in that respect. Going multi-threaded is quite easy generally speaking for image-based work (you can divide the image in several pieces and work in parallel), and so the more cores you can get, again, the better it will be.

That said, programming the PS3\'s eight Synergistic Processing Units (SPUs) can be quite challenging from time to time, but there is definitely a lot to gain from the architecture, and ProFX on PS3 should run impressively fast. (We should get confirmation on that by the end of the year or so

A very cool side effect of the PS3\'s architecture is that ProFX should be able to continuously stream textures -- using one or two SPUs to compute the textures to be given to the GPU to be displayed, that\'s a huge bonus for adding rich content to games without having to stream that content from the BluRay or overloading the sole GPU of the machine.

You should see more graphically impressive games than ever with that kind of combination.

DDJ: We\'ve seem to have made great strides in terms of texture mapping and other forms of 3D modeling. What\'s the next big challenge?

SD: There are many technical challenges that I could point out here. But I\'d like to focus on one. Game developers now have a lot more content to produce, store, and distribute than ever before. MMORPGs are a good example here, filling up a BluRay of 50GB of content (instead of 5GB for a single layer DVD) is another one.

The question is then, for producing 10 times more content, do you want to grow your company by a factor of 10? Although I like the idea of allowing the industry to grow significantly, and although I love the idea of providing the maximum people with a cool job, I don\'t believe it\'s feasible that way.

Therefore game developers have to find new ways of not only producing that content, but also storing and delivering it.

I strongly believe procedural techniques for content creation are a major solution to that challenge:

* It\'s proven that, because defining procedural content is a highly non-linear process, you produce more content faster than with traditional techniques by reusing a lot of the base elements you\'ve been producing along the line.
* Procedural content can be easily customized and created by gamers, and user-generated content is another key element for producing never-ending content.
* Procedural content files are typically parameters or code, which is usually extremely small compared to what it can produce. And that makes them ideal for the storage and furthermore the delivery of that content.

Of course, going from a traditional approach for content production to using generative techniques can be quite a challenge by itself, but it\'s worth the effort, and one of the only ways I see for raising the bar in terms of next-generation content creation.

DDJ: If readers want to find out more about these topics, can you suggest a web site? [okay to point to http://www.allegorithmic.com - YES]

SD: First, I invite readers to take a look at the web site dedicated to ProFX, Allegorithmic\'s middleware for procedural texturing, where they can find a lot of information regarding procedural texturing for games:

They can also check the Wikipedia webpage about procedural generation as a whole, and see that it\'s quite a hot topic right now (waiting for Spore!

http://www.neogaf.com/forum/showthread.php?t=184843&page=10


Interesting reading !

Cliff notes or no ready.

[BizioEE]

Source: n4g.com

DDJ: Sebastien, the Allegorithmic web site mentions "procedural textures." What does that mean?

SD: Procedural textures are textures defined according to an algorithm. Instead of painting (by hand) the texture pixel by pixel, you define the way these pixels have to be lit to produce the texture you want. It\'s basically like writing a program, and with that program comes all the power of algorithms over frozen (bitmap) data -- compactness, data amplification, parameterization, and the like.

When the procedural texture is defined, you launch an engine for actually generating the bitmap textures. These are called realizations of the procedural description, and it\'s a lot like the difference between a -- possibly random -- process and the realization of that process.

In the case of Allegorithmic\'s procedural textures, you\'d define the procedures in a nice visual way using our product MaPZone, then use our other technology ProFX to rasterize the images, producing the final result to be displayed by a game engine or any renderer.

There are many advantages to using procedural textures instead of bitmap textures, one of them being that procedural texture files are typically 500-1000 smaller than bitmaps (making them ideal for online games).

DDJ: I\'m familiar with 3D textures. But what are 4D textures?

SD: 4D here stands for 3D + Time. 4D textures means evolving textures, evolving environments and games. On our web site we showcase a demo around that idea, where a bathroom goes from a clean state to a dirty state in real-time.   Mov1 http://download.profxengine.com/gallery/demos/bathroom/videos/bathroom_01.mov  
Mov2 http://download.profxengine.com/gallery/demos/bathroom/videos/bathroom_02.mov

Procedural textures, because they are driven by parameters, can be driven by time. And by letting time go by, the aspect of the textures can be changed, going for example from a clean state to a deprecated one.

Now imagine that applied to a game, where the whole environment and characters are changing, evolving, according to time but also to your actions within that environment. Actual PCs and consoles can handle this, and I must say I can\'t wait to see games using this as a feature.

DDJ: If I recall from your web site, a platform you will shortly be supporting is Playstation 3. What benefits and challenges do multi-core processors bring to the scene?

SD: Procedural techniques are demanding in terms of power, so the more power you can get, the better it is in that respect. Going multi-threaded is quite easy generally speaking for image-based work (you can divide the image in several pieces and work in parallel), and so the more cores you can get, again, the better it will be.

That said, programming the PS3\'s eight Synergistic Processing Units (SPUs) can be quite challenging from time to time, but there is definitely a lot to gain from the architecture, and ProFX on PS3 should run impressively fast. (We should get confirmation on that by the end of the year or so.

A very cool side effect of the PS3\'s architecture is that ProFX should be able to continuously stream textures -- using one or two SPUs to compute the textures to be given to the GPU to be displayed, that\'s a huge bonus for adding rich content to games without having to stream that content from the BluRay or overloading the sole GPU of the machine.

You should see more graphically impressive games than ever with that kind of combination.

DDJ: We\'ve seem to have made great strides in terms of texture mapping and other forms of 3D modeling. What\'s the next big challenge?

SD: There are many technical challenges that I could point out here. But I\'d like to focus on one. Game developers now have a lot more content to produce, store, and distribute than ever before. MMORPGs are a good example here, filling up a BluRay of 50GB of content (instead of 5GB for a single layer DVD) is another one.

The question is then, for producing 10 times more content, do you want to grow your company by a factor of 10? Although I like the idea of allowing the industry to grow significantly, and although I love the idea of providing the maximum people with a cool job, I don\'t believe it\'s feasible that way.

Therefore game developers have to find new ways of not only producing that content, but also storing and delivering it.

I strongly believe procedural techniques for content creation are a major solution to that challenge:

* It\'s proven that, because defining procedural content is a highly non-linear process, you produce more content faster than with traditional techniques by reusing a lot of the base elements you\'ve been producing along the line.
* Procedural content can be easily customized and created by gamers, and user-generated content is another key element for producing never-ending content.
* Procedural content files are typically parameters or code, which is usually extremely small compared to what it can produce. And that makes them ideal for the storage and furthermore the delivery of that content.

Of course, going from a traditional approach for content production to using generative techniques can be quite a challenge by itself, but it\'s worth the effort, and one of the only ways I see for raising the bar in terms of next-generation content creation.

DDJ: If readers want to find out more about these topics, can you suggest a web site? [okay to point to http://www.allegorithmic.com - YES]

SD: First, I invite readers to take a look at the web site dedicated to ProFX, Allegorithmic\'s middleware for procedural texturing, where they can find a lot of information regarding procedural texturing for games.

http://www.profxengine.com/index.php?PAGE=GALLERY.DEMOS

http://www.neogaf.com/forum/showthread.php?t=184843&page=10


Interesting reading !

[Paul2]

lol, bizioee.  who\'s going to sit through that and read it?

I started to...then I fell alseep during it.

[BizioEE]

It was just some info/tech related to "improvement between gens"...

[BizioEE]

For the ones who are interested !
A lot of info about PS3 hardware, specs, Cell, RSX, Bandwidths and the huge potential of PS3.

Part 1

RSX Data Rides a Fast Bus to XDR Memory on the PS3\'s!

READY TO TAKE A RIDE? ....... on a Bullet Train of information?
I\'m going to make an attempt at correcting all the  misconceptions, misinformation and misdirection we all have been pounded with over the last 3yrs on the PS3 Bandwidths, specs, mis-labeled boards, withholding of information by Sony the Media and Silent Developers with NDA restrictions!

Here\'s the Rambus Press Release/specs specifically for PS3 as of Nov \'06:
 [reveal][/reveal]*****
XDR Memory Modules have Dual Bandwidth data rate of  6.4GB/s @3.2GHz Bandwidth per module = 25.6GB/s @3.2GHz  and  8GB/s @ 4.0GHz for a Total of 32GB/s TOTAL XDR Memory Bandwidth Available to both the Cell and RSX!

[reveal][/reveal]*****

The XCG Memory Clock Generator balancing and synchronizing the frequency of the Dual Bandwidth capable 256MG of XDR Memory Modules on PS3. This handles the speed of memory access of both the Cell and RSX. While the Dual XiO Controllers on the Cell handle the Bandwidth control!
[reveal][/reveal]*****

FLEXPHASE Bus Interface: Dual Controllers on Cell\'s XiO Interface enable two separate data rates of 25.6GB/s @ 3.2GHz (6.4GB/s per module @3.2GHz) and 32GB/s (8GB/s per module @4.0GHz) of Aggregate Bandwidth on a Data Bullet Ride accelerated by RAMDACS  .  So Peak Bandwidth over FLEXPHASE of 32GB/s to the RSX is possible!  XCG (Clock Generator) controller is what is responsible for setting these rates over the Dual XiO Memory Controllers. It can synchronize the CLOCKS for all FLEXPHASE interfaces. The dual XiO controllers on Cell simply control their Bandwidth signaling. So it is Flexible Bandwidth interface capable of switching between 25GB/s and 32GB/s (if the Memory is capable of that, which it is according to Elpida at either 6.4GB/s or 8GB/s per module)!

[reveal][/reveal]*****

FLEXiO Interfaces offer flexible data speeds between 400MHz - 8GHz with fully scalable over two configurable interfaces with Total Total Bandwidth Access of  76.8GB/s @ 6.4Gbps!  ........this includes over 60GB/s Total Bandwidth to RSX split 25GB/s from RSX and 35GB/s @ 6.4Gbps at up to 8GHz. The other interface is to the Super Companion chip. For the remaining 16.8GB/s @ 6.4Gbps of that 76.8GB/s Total Bandwidth! (this is scalable with the variable speed of  between 400MHz and 8GHz)

[reveal][/reveal]*****
This is a diagram of the on chip memory Bandwidths with all three of Rambus FLEX Interface System connections. Notice Cell\'s XiO to XDR Memory and on the FLEXiO side, we have two separate connections of Cell\'s iO interface. Which is connecting to both the Main iO Interface of the RSX and the separate interface to RSX\'s Memory Bus. Where Cell is capable of streaming from GDDR3 Memory and to the RSX and it\'s Memory directly!
[reveal][/reveal]
While we\'re here we will look at two other interfaces off the Cell on this chart. They include the BIF (Bus Interface) and XiO interface to the RSX, and remember these are fully programmable and scalable as per Rambus specs on 33 visible double lanes of the Mother Board. They are accelerated by the Ramdacs beyond the speed of the chip sending the data signals. Much like a slingshot across the Bus from as little as 400GHz to as much as 8GHz. I hope you understand that the entire FLEX interface system is not locked (like traditional Bus interfaces) and uses the both iO and XiO (dual controller) controllers on the Cell to scale bandwidth according to system needs.

 Look in that Diagram!
Now what do you think that BiF interface is for in the diagram above??? .......well I\'ve alluded to it before. It connects the Cell EIB to the RSX Memory Bus Interface. So that RSX not only has the 4 x 7.2GB/s with a Total Local Memory Bandwidth of  28.8GB/s with it\'s GDDR3 (22.6GB/s was for old specs on lower grade memory), but along with that, straight through the programmable scalable BiF (via  FLEXiO Bus Interface), we have XDR access of 25.6+ GB/s (not fixed)! So lets just say we now have potentially faster write access to the XDR Memory than the RSX\'s Main Memory of  256MG GDDR3 at 28.8GB/s for a Total Aggregate Bandwidth of OVER 54.4GB/s of available Graphics Memory Bandwidth. The Peak for RSX would then be a Total of 60.8GB/s Possible! (of course this isn\'t taking into account what the Cell is using, but it is all dynamically balanced and synchronized. So they are hypothetical maximums that naturally always keeps whatever the Cell needs taking priority over RSX\'s demands on it\'s XDR Memory)

Note: Samsung GDDR3 Specs, is where we get  32 x 8 bit rate = 256bit rate (not 128bit as some boards say). Bandwidth 7.2 x 4 =  28.8GB/s out of all modules not 22.4GB/s of standard DDR on some of those E3 2005 Boards.

Now that means the CrAzY insane possibility of the RSX not being an off the shelf G70 Series is not only possible but completely destroys this "Myth". Is it Based on G70 Series Programmable Shader Technology? Yes!  Because when you add the fact that any G70/G71 Series GPU could not handle the FLEXiO number of pin connects and speed in a million years, maybe you\'ll understand RSX must be a Custom ASIC Design (no off the shelf G70, it\'s impossible, you can add pins where they\'re aren\'t any ). We are talking a Uniquely Designed GPU,  to fit these kinds of specs. Mainly the PS3\'s ability to push  4K Resolutions  to the HDMI connection to a Screen.  

(Note: At this point this requires a Double-Link HDMI Connection to a Screen and won\'t be possible till Sony adds another HDMI connection in the Future!)
---------------------------------------------------------------------------------------------------------------------------------------------------------

[BizioEE]

Part 2

PS3 is Alive with New Innovative eFuse Technology.

Our body controls our functioning without us being aware of it. We get hot, it cools us down. We sweat. Things go wrong in our system and our body goes to fix it. It regulates it own pulse and presure. Now the ability of the PS3 to dynamically clock itself is done in very much the same way.

Here\'s why! Over 3 years ago IBM designed eFUSE Technology. It is now used in not only the Cell, but their Power6 chips as well. Inside the cell are some technologies that run an integrated engine within the Cell and regulates all of it\'s functions automatically. Two of those elements of technology needed to run Dynamic Self Clocking are eFUSE (check link it\'s Publicly available information) and 10 Thermal Sensors. Two on the PPE and one for each SPE. If you look at the picture below, you will see heat patterns on the Cell BE (also notice that all cores are active to some extent and this is from the PS3 version of Cell. The two far cores from the PPE show less activity, but none the less they are not hardware disabled or they would not show any activity.
[reveal][/reveal]*****
So a dynamic clock is one that is scalable in response to multiple feedback types such as workload, temperature, etc, for every single core on the Cell BE. The engines on the Cell that are responsible for running this system, take all feedback into account and send energy to these eFUSE elements on each core(and they are like a resistance fuse that is dynamic from blown to full contact). Thereby varying the voltage to each core which in turn effects it\'s frequency clock speed! .....could this also be the case with the RSX? ooh.... I wonder. I think it\'s interesting that with this type of mechanism in place on Cell that 1 core could be just sharing it\'s cache or PPE could be left running at 3.2GHz and and individual cores could be clocked much higher. One starts to get hot the work is switched to another core or rotated to several other cores. For a short time you could hit the trottle and all SPE\'s out. But then you realize it\'s doing it all itself and instinctively knows how to keep cool and pump power through the entire PS3 system (like how your blood cools and feeds your own body)! OMG it\'s the uber... chip come alive, run for your lives before it eats us all. LOLz

It has been said, that Cell seems to enjoy running between 3.2 to 4.0GHz. But in F@H even with the Cell that\'s in the 60gig it\'s been observed running at over 5GHz. I can even show you a Sony board with it running at 4.0 on F@H. So just because I said the Cell is capable of running over 6GHz, don\'t assumed I was talking about a fixed clock at that speed. I\'ve the said "Dynamic Self Clocking" in Real Time in response to heat, function corrections, and computational demands, etc. Now we know how it does it:

Here we have a slide on the Cell on PS3 showing a CPU Clock of 4.0GHz! ...notice it says, Digital HOME and Distributed Computing.
[reveal][/reveal]*****

Quote From IBM\'s page on eFUSE:
Quoted from 2004 on IBM\'s announcement of "IBM "eFuse" tech will make chips self-configuring"
 Within this post are some of my over 4 years of Cell BE and PS3 Research. Attempting to pull the pieces together. There is a lot of information here for you to digest. Hopefully you are not overwhelmed. If you can\'t handle it, then just forget about reading this and enjoy your games whether that is on a PS3, a PC, the Xbox 360 or the Nintendo Wii. But remember that the people who worked the hardest to make sure you have a complete media center and game console that is capable of lasting well into the next decade is once again Sony with your PS3! :mrgreen:  ......Enjoy

Much Thanks goes to the Engineers at IBM, Rambus, Nvidia, Samsung, Toshiba, NEC, Sony, Marvel, and the 1000\'s of lives dedicated to this engineering masterpiece.

Taken from PS3forums.

[Viper_Fujax]

why not just look forward to the ps4 instead of the hopeful potential of the ps3?

at least with the ps4, you know it will definitly be better than this generation right now..

[BizioEE]

why not just look forward to the ps4 instead of the hopeful potential of the ps3?

at least with the ps4, you know it will definitly be better than this generation right now..

You already have the first result = R&C, Uncharted, GT5 prologue, COD4, DMC4, Burnout Paradise, etc and will see improvements year by year, why you should look forward to the PS4...
...as I told some time ago it\'s just tech/info for tho ones interested...that\'s all...in a thread devoted to "improvement between gens"...

Part 2



Taken from PS3forums !

There is something I dont get on the last slide.

AFAIK there are no 9 cores in CELL. Is it really reliable? Are we sure this is indeed for the PS3\'s CELL??

[Viper_Fujax]

You already have the first result = R&C, Uncharted, GT5 prologue, COD4, DMC4, Burnout Paradise, etc and will see improvements year by year, why you should look forward to the PS4...
...as I told some time ago it\'s just tech/info for tho ones interested...that\'s all...in a thread devoted to "improvement between gens"...

yea i just hate exclamation points. of which you\'re the king of

[Paul2]

There is something I dont get on the last slide.

AFAIK there are no 9 cores in CELL. Is it really reliable? Are we sure this is indeed for the PS3\'s CELL??

not only that, but even i don\'t about the 4 GHz speed.  Isn\'t the PS3 only 3.2 Ghz?