nVidia building the PS3 GPU in its "entirety"

[London Geezer]

PS. While I'm without a doubt an ass (no pun intended), can we ditch the link to Deadmeat's picture? Thanks.

What link are you talking about??

Link dutifully ditched.

 

[Qroach]

What WTF are you talking about?

 

[Qroach]

if he thinks I've been linking to deadmeat, or anything that idiot says, he can think again!

 

[London Geezer]

if he thinks I've been linking to deadmeat, or anything that idiot says, he can think again!

Nooooo... there was a link in my post with a picture, it was Hey69's link i fixed for him, so he could copy it and paste it to his sig. It had a very nasty looking picture.
Now i took it off.

 

[Sonic]

Bring the thread back on topic.

 

[one]

x86 isn't getting SPUs because its doesn't need them, it will use seperate GPUs for when you need lots of floating point maths.

Yeah x86 is unchanged (or changed a bit in x86-64), but the world is changing. What the people in the world want in computers is changing. Without this change in demand, there was no 'multimedia'-focused enhancement in the PC such as NetBurst. Quoting the exact comment of KK from the press release,

"Massive and rich content, like multi-channel HD broadcasting programs as well as mega-pixel digital still/movie images captured by high-resolution CCD/CMOS imagers, require huge amount of media processing in real-time. In the future, all forms of digital content will be converged and fused onto the broadband network, and will start to explode," said Ken Kutaragi, executive deputy president and COO, Sony Corporation, and president and Group CEO, Sony Computer Entertainment Inc. "To access and/or browse sea of content freely in real-time, more sophisticated GUI within the 3D world will become the 'key' in the future. Current PC architecture is nearing its limits, in both processing power and bus bandwidth, for handling such rich applications."

 

[Vince]

No Sony is embracing a version of the PC architecture.
Cell is hoping to become the next x86. A standard that works o.k. in most cases.

You think? I, personally, don't see Cell advancing as x86 did. I wonder if we'll see such open and differentiated development happening around it as we do between companies such as AMD and Intel and Cyrix. But, none of us can answer that so...

I do disagree on the standard that works O.K. comment. I think it's pretty clear that at the core it's targetted at a particular area of Computation/Bandwith heavy applications (harking back to Diefendorff and Dubey) that will work O.K. when tasked with other things whose resource requirements scale alot alower or have basically stopped for all intents and purposed. This being said, I recognize that many things assocated with gaming don't mesh well, namely the fixed-functional back-end.

All thats really new is fitting more advanced vector units than usual for a general processor. What I'm saying is that there isn't yet an inherit advantage of Sony's stragegy (except to Sony) than the normal IT industry (which uses x86, PowerPC or ARMs as a rule).

The inherient advantage is holistic, it's in the preformance delta between Sony's PS3 console and the PC at the time it comes out. Do you think the PC will be comparable, I sure as shit don't.

x86 isn't getting SPUs because its doesn't need them, it will use seperate GPUs for when you need lots of floating point maths.

And back to my origional argument we go. If you could do away with the arbitrary divides between vendors and standards and the open platform and the industry was entirely vertical, would we have the current CPU + GPU paradigm as you just articulated?

I'd say no, and I'd think Microsoft agrees with me if X2 is any indication of a trend.


And, please, you don't need to go after the revolutionary speel. I'm not coming from that direction, I'm not saying it is. I'm saying it's going to cure world hunger and get deadmeat laid; I'm just saying it's a departure from the PC and it's significant in it's holistic view towards a computing system and, beyond that, I think it'll have alot of potential for digital content distibution and other non-gaming but still important for Sony tasks.

 

[pahcman]

KK is complelety wrong when he says PC architecture has run out of steam...

KK said the same think with PS2

MrSingh also said Cell is just PS2 on steriods. He has his japan sources, first to break the Nvidia news, right Panajeva and Fafalada?

Just think of SPU's as VU's on steroids.

In fact, the best way to describe Cell is to think of it as a PS2 on steroids. (1 + 2 -> 1 +8)

Panajeva share with us how much more info you manage squeeze out lately?

 

[London Geezer]

Isn't any new architecture the "last one on steroids" plus some new features?

 

[one]

KK is complelety wrong when he says PC architecture has run out of steam...

KK said the same think with PS2

MrSingh also said Cell is just PS2 on steriods. He has his japan sources, first to break the Nvidia news, right Panajeva and Fafalada?

Just think of SPU's as VU's on steroids.

In fact, the best way to describe Cell is to think of it as a PS2 on steroids. (1 + 2 -> 1 +8)

Panajeva share with us how much more info you manage squeeze out lately?

Hello? Are you OK?
PS2 succeeded commercially. What's wrong with that?

 

[Dave Baumann]

• I have GPU Gn, CPU Cn, Storage Sn, which of the following would more closely approximate the eq. point of greastest preformance:
  [list:0901d49371]
• Closed System A: [GPU G1, CPU C1, Storage S1] with custom interconnections and system optimization; or:
• Open System B: Random combination of [GPU Gn, CPU Cn, Storage Sn] |n=1...1E5

[/list:u:0901d49371]
The PC will allways lose out to a closed-set enviroment, especially in the PC paradigm we have in which there is little thinking of the system on a holistic level. Each vender is concerned primarily with their given component and there is little cooperative work. Hell, look at the history of PCI, AGP and - finally - PCI-Express. AGP texturing anyone?

Actually I fail to see how currently available PC inteconnect devices will fundamentally differ from anything in closed box systems of a similar size/performance/ilk within a similar timeframe right now. You may talk derisively about AGP texturing but that was a technology development many, many years ago (and it still being used) - what type of graphics were closed home box home entertainment devices producing at the time? The PC indsutry has moved on to the PCI Express interface, removing the legacy issues behind a parallel interface - what "closed computing systems" are currently, right now, on the market that matches it? In what ways is this not pushing technologies available today to a high degree and in what way does this mean that other steps won't come in similar timeframes to other competing technologoes used elsewhere? The PC has evolved to suite the very changing needs of its environment and the techologies it employs will continue to evolve along the same lines - JHH believes this to be the case as well as he pretty much stated it as well.

And, no, its not "finally" PCI Express, its "currently" PCI Express.

For example, as the ATI guys have told me and you've mentioned on the site from time to time, general computation is moving to the GPU. Not to mention names, but somone here and I were discussing sequencing on them. A GPU is, I'm betting you're going to say, highly tuned to graphic applications but it can still run anything.

And it is, thats the point - we've mentioned in serveral threads about providing much functionality for hiding texture latency as a clear example of that. The other point being is that we are seeing CPU functionality move to the graphics processor, as yet we're not talking about it going the other way.

 

[Vince]

Vince could you please give your definitions for "PC paradigm" and "Cell paradigm".

Sure, I don't know if labeling it a 'Cell paradigm' is appropriate, bit I was referring to the difference between an open-set of potential combinations taken from a multiple vendor pool that's dictated by market dynamics (PC-esque) and a model which is a closed-set and holistically designed as a closed-set (Cell-esque).

While the open system is advantaged when it comes to reaching a low price point, I've questioned how closely it approximates the region of optimum preformance. Following from here, it's clear that a holistically built device (component invarient for argumetn sake) will have certain intrinsic benefits when it comes to preformance over an open system.

I feel like I'm repeating (because I am), but of course in a competitive enviroment a closed-set will lose out due to marketplace dynamics. Which is why the CE industry and, in particular, the Console field is so unique, it's inheriently closed and it's a good catalyst for a move into the CE field. Sony is even more unique as they are a company that can make this work, controlling the entire pipeline from front-end content creation to the back-end CE devices, they can use an architecture like Cell and Broadband to create synergistic value for the consumer.

 

[pahcman]

Sure PS2 succeed as a simple game console, like others before it.

The inherient advantage is holistic, it's in the preformance delta between Sony's PS3 console and the PC at the time it comes out. Do you think the PC will be comparable, I sure as shit don't.

I dont get it, if PC has reached limit, and this Cell revolution will remedy these limits, then shouldnt the Cell advantage be persistent? Why stop a the time it comes out? Shouldnt PS3 process undisputed displayed power till the next PS4 revolution?

One thing i sure PS3 will not beat PC at its time, is more RAM more Storage more Functions.

 

[Panajev2001a]

regarding EDram: since Sony is goning to fab the gpu, nvidia is obviously designing it using their design libraries.

SCE's and Toshiba's design libraries ? That would make sense, but they started the design with other tools so the transition process to the ST (Sony/SCE-Toshiba) design libraries will take part of the development time of the custom GPU.

I think 50 of nVIDIA's best engineers (major NV50 work has prolly finished already and they can re-use what they develop working with SCE on the NV60 or try to converge the R&D on the custom parts of NV5X-PS3) engineers should be able to deliver quite a fast architecture.

I think that nVIDIA and SCE can adn will collaborate quite well together: they both are competent designers of complex ICs with a good amount of skilled engineers at their service.

They mention how they see the architecture they are developing for/with Sony/SCE as being present in a multitude of CELL based devices (he mentions how the whole PlayStation 3 platform would become the basis for most of Sony's CE products [scaled according to the product's needs of course]).

I think this extends beyond simply taking the chip and disabling "quads" or removing "quads" from the chip: I think when they designed NV5X they gave a hard thought about the whole range of devices they would want to see the architecture be present in. I think they understood that their goals and what Sony/SCE wanted were not that different from each other.

I think nVIDIA did win the GPU contract because they managed to convince Sony/SCE's management that the future was Shaders, but also because the philosophy of their next-generation architecture went along quite well with the philosophy Sony/SCE had in regards to the CELL architecture.

 

[Vince]

Actually I fail to see how currently available PC inteconnect devices will fundamentally differ from anything in closed box systems of a similar size/performance/ilk within a similar timeframe right now. You may talk derisively about AGP texturing but that was a technology development many, many years ago (and it still being used) - what type of graphics were closed home box home entertainment devices producing at the time?

Dave, ask a console developer here if they have greater flexibility in using resource sharing and balancing processing between components on the PC or on a console.

The PC indsutry has moved on to the PCI Express interface, removing the legacy issues behind a parallel interface - what "closed computing systems" are currently, right now, on the market that matches it? In what ways is this not pushing technologies available today to a high degree and in what way does this mean that other steps won't come in similar timeframes to other competing technologoes used elsewhere?

Dave, any open-system will need standards; and standards that are mediated by committee will be slower that what a single vertical organization can do. For example, if Sony wants XDR or RaZor, it gets it. AGP was launched in what, 1998... so it's been 6 years or there-abouts?

Blue-Gene's interface blows it away AFAIK. As will Cells in a a few months.

And it is, thats the point - we've mentioned in serveral threads about providing much functionality for hiding texture latency as a clear example of that. The other point being is that we are seeing CPU functionality move to the graphics processor, as yet we're not talking about it going the other way.

Wow! Now do you get why I've been asking you what seperated an X2 ALU from an APU core? Cell isn't built around the PUs Dave... at it's heart it's a large concurrent SIMD Vector processor in the same vein as a unified shading architecture, but a hell of alot faster in clock and more flexible.

Secondly, an SPU complex addresses these concerns to some extent and the basic concept can be extended. That's what you're not understanding, the SIMD cores are very close (Faf answered for you), the difference is between the complex built around the computational resources (eg. SPU or in the Shaders). AND, we don't know fully how their DMAC system works and how the PU/DMAC arbitrate. Your comments are premature and, likely, will come around to get you.

 

[Tuttle]

I think nVIDIA did win the GPU contract because they managed to convince Sony/SCE's management that the future was Shaders, but also because the philosophy of their next-generation architecture went along quite well with the philosophy Sony/SCE had in regards to the CELL architecture.

Future is shaders? I don't think so. PC shaders are nothing more than an artifact of the hardware layout of desktop computers.

 

[Vince]

Future is shaders? I don't think so. PC shaders are nothing more than an artifact of the hardware layout of desktop computers.

Someone just touched the third rail... do I jump on or not?

 

[Panajev2001a]

The XBox shows this is true, and refutes your dumb earlier post defending the current model on the grounds that it's prolific (totally unrelated to the argument at hand), by showing that the same components when taken out of the competitive PC paradigm will outpreform it.

Its not dumb Vince, your statement that the PC vendors have an "(in)ability to create usable and farsignted standard between each component" is clearly incorrect as you have proved yourself by the very fact that it can be applied beyond the scope of the PC environment and used to a full context.

Dave, look at the NV2A, look at the NV20... I think that some times there are clearly cases in which nVIDIA (or ATI) saw performance best attainable one way, saw the future being in a certain direction, requiring certain customizations... yet, let's say, Microsoft (being lobbied by ATI and other companies other than just nVIDIA) disagreed and pushed DirectX in another direction.

Under DirectX, nVIDIA is limited in what they can expose of their innovative ways and if OpenGL for gaming went *puff* and DirectX were to be the main and basically only commercially viable solution in the PC space then you would see IMHO a model that would have to follow Microsoft's pace and not the industry's pace.

The PC model tends also to prefer ease of implementation and ease of backward-compatibility to performance (compatibility in general takes a front-seat compared to performance) as indicated once again by the industry as they chose the slowest of the three standards proposed for PCI-Express 2.0 (5.0 Gbps).

A console, something like PlayStation 3 or Xbox 2/Xenon, not being a Desktop PC and not having to be as compatible as another PC made by Dell or HP, but being its own closed platform can afford to do much more than simply mixing and matching off-the-shelf PC parts with no customization.

Look at the Xbox 2 architecture or a bit more in the future the PlayStation 3 architecture: they run custom software, developed for highly customized hardware (that does not need to employ the technologies which are available and are used in Desktop PCs) and customized APIs (DirectX for Xbox 1's inner-workings were completely re-written and optimizeed for the Xbox 1 architecture and specific customizations present in the NV2A were uncovered... customizations that DirectX on PC would have not uncovered).

 

[Joe DeFuria]

What is most interesting to me is how Jen-Hsun commented again that this will be used in all Sony CE devices, which is what Cell is intended and was designed for -- We have independent confirmation from both Sony group, IBM and Toshiba on this. So, how would one make these statements logically compatable?

Don't get your question.

Why is this interesting, any more than ATIs and nVidia's already known plans (and already existing products) of pushing their GPUs into other consumer devices, like TVs, set-top boxes, cell phones, etc?

 

[Panajev2001a]

Of course a closed design can win the short term but lose in the long term. Indeed STI know this, by trying to get Cell into into its own market economy model.

So, you spend most of the post dissagreeing and then undermine it all by agreeing with me in the short-term. Which is all that's relevent to this discussion? *confused*

No Sony is embracing a version of the PC architecture.
Cell is hoping to become the next x86. A standard that works o.k. in most cases.
Moving from a custom designed processor for every generation to using a basically off-the-shelf part thats compatible with the lots of already written software.
All thats really new is fitting more advanced vector units than usual for a general processor.
What I'm saying is that there isn't yet an inherit advantage of Sony's stragegy (except to Sony) than the normal IT industry (which uses x86, PowerPC or ARMs as a rule). Sony is hoping for one processor to rule them all, but I don't see any reason why Cell will do any better than any other plan for global domination.

x86 isn't getting SPUs because its doesn't need them, it will use seperate GPUs for when you need lots of floating point maths.

I fail to see why you see Cell as so revolutionary, its just a CPU with a few independent vector units.

We agree it is not a revolution, but an evolution of the existing theories and practices in micro-processors' design it is.

I really dislike this... bah it takes a general processor and slaps on top of them some vector units.

This is overly simplicistic and off-the-spot IMHO: hey, ya know what... there is not much difference between a BMW M5 and a Yugo... I mean one just has a fast engine slapped on, but both have a chassis, 4 wheels, suspensions and seats... the same car...

If you break down the basic ideas that CELL was designed with... you'll see that they are not brand new... they have been theorized and experimented mostly "separately" for a long time in the industry.

I see CELL as an architecture that finally, thanks also to the advances in manufacturing technology that make this possible, brings all those ideas together and glue them quite efficiently.

It is like a sport team (of any sport really): the winning team does not just have great players, it has a great coach that knows how to keep the team playing the same game efficiently and see how things should be arranged on the field to allow each player to perform at their best.

It is like spitting on BlueGene/L systems or Itanium 2 based systems like SGI's Columbia because they are based on general-purpose MPUs: go build a comparable system... I bet that after you make the massive newegg.com order for the processors that you will have to sit with tons of empty boxes as you miss the foundations on which those processors rely on and that allows them to communicate, reach data and share data at the highest possible speeds.

I see CELL as an architeture in which each major component was thought and designed both independently (we want a fast and advanced Vector/Scalar unit design) and as parts of a CELL based processor (What does CELL need ? How can we maximize the whole MPU's performance ?).