RSX patent?

[BlueTsunami]

But wasn't the Xbox itself very close to the PC architecture? Meaning, isn't it kinda hard to compare the NV GPU in the Xbox and the NV GPU in the PS3 because the architectures are very different from eachother? Also...no one is claiming its nothing OTHER than a G70 paki you keep bringing it up as if someone is.

EDIT: Or am I completley wrong with the Xbox having a that type of architecture? I'm probably buying into the myth, even though I don't know better.

 

[pakpassion]

But wasn't the Xbox itself very close to the PC architecture? Meaning, isn't it kinda hard to compare the NV GPU in the Xbox and the NV GPU in the PS3 because the architectures are very different from eachother? Also...no one is claiming its nothing OTHER than a G70 paki you keep bringing it up as if someone is.

im not talking about the CPU architecture. im talking about the way the GPU-CPU interacted (minus the flexIO).

 

[BlueTsunami]

im not talking about the CPU architecture. im talking about the way the GPU-CPU interacted (minus the flexIO).

Ok then, well compare the CPU in the Xbox and the CPU in the PS3. I would consider that a radical change or no?.

 

[pakpassion]

Ok then, well compare the CPU in the Xbox and the CPU in the PS3. I would consider that a radical change or no?.

all next gen consoles = radical change

 

[BlueTsunami]

all next gen consoles = radical change

Yes I know, but i'm talking about in relation to the relationship between the GPU and CPU. There must be a radical change between the GPU in the XBOX and how it interacts with the CPU in the PS3. How they communicate is not the same. I'm trying to say that it must be different from how its done on the PC. Not only to have it work with CELL in the first place but to possibly benefit from being closer (remember teamwork!) to eachother.

Your trying to say that the GPU in the PS3 is similar to the GPU in the XBOX in that their JUST supped up GPUs from the same class. As we progress foward, finding out more and more on how they interact (Cell and RSX) its begining not to look like a supped up G70 but something else that has the same arcitecture of the G70 but it doesn't mean its the same situation with the XBOX.

EDIT: I would like to point out that I have no idea what i'm talking about...

*runs away

 

[pakpassion]

Yes I know, but i'm talking about in relation to the relationship between the GPU and CPU. There must be a radical change between the GPU in the XBOX and how it interacts with the CPU in the PS3. How they communicate is not the same. I'm trying to say that it must be different from how its done on the PC. Not only to have it work with CELL in the first place but to possibly benefit from being closer (remember teamwork!) to eachother.

Your trying to say that the GPU in the PS3 is similar to the GPU in the XBOX in that their JUST supped up GPUs from the same class. As we progress foward, finding out more and more on how they interact (Cell and RSX) its begining not to look like a supped up G70 but something else that has the same arcitecture of the G70 but it doesn't mean its the same situation with the XBOX.

Im just saying there is some discrepency in people might believe that RSX might somehow access the SPEs of Cell but the truth is, RSX will be able to access System memory like a PC GPU does but like the Xbox the GPU and CPU will work together and share the load. That is a standard of every closed system console. Its not as if Xbox 360 and revolution wont do it. They both will. especially in Xbox 360 where the GPU can make the core or one of its threads a Slave and use that slave to do the GPU's work.

 

[Shifty Geezer]

im not talking about the CPU architecture. im talking about the way the GPU-CPU interacted (minus the flexIO).

He's given no info at all about this but we know there's other things going on, like a dedicated 35 GB/s communications between the CPU and GPU which is there to share data, both ways. We know from other discussions that it's theoretically possible for tiles to be sent to Cell to work on for example, or graphics data to be fed directly to Cell for physics processing. The interaction between Cell and RSX is going to be a lot more than a typical PC-like CPU-feeds-GPU structure. A few lines in a brief Q&A in a non-technical fanzine isn't likely to throw out more than just marketting spiel, and isn't likely to be carefully worded. We can dissect statements like these in a number of different ways. eg.

The RSX is designed to be a powerful, efficient companion processor to the Cell.

Doesn't that imply GPGPU functionality? If one wants to interpret it as such, sure! The whole answer to Question 3 (and 1 and 2 for that matter) is not a very cohesive statement with clearly defined points Jen-Hsun Huang is trying to convey, as these articles rarely are. I think you're trying to get more information out of the article than is in there.

 

[pakpassion]

He's given no info at all about this but we know there's other things going on, like a dedicated 35 GB/s communications between the CPU and GPU which is there to share data, both ways. We know from other discussions that it's theoretically possible for tiles to be sent to Cell to work on for example, or graphics data to be fed directly to Cell for physics processing. The interaction between Cell and RSX is going to be a lot more than a typical PC-like CPU-feeds-GPU structure. A few lines in a brief Q&A in a non-technical fanzine isn't likely to throw out more than just marketting spiel, and isn't likely to be carefully worded. We can dissect statements like these in a number of different ways. eg.
Doesn't that imply GPGPU functionality? If one wants to interpret it as such, sure! The whole answer to Question 3 (and 1 and 2 for that matter) is not a very cohesive statement with clearly defined points Jen-Hsun Huang is trying to convey, as these articles rarely are. I think you're trying to get more information out of the article than is in there.

first its 20 GB write, 15 gb read, not 35 both ways as some might think. Theoretically EE did alot of things that still havent been achieved so lets base it on real information. Physics Processing is done mostly through the Cell just like in Xbox 360 and RSX or Xenos in 1st gen games will hardly ever take part in any Physics detailing. Its true that the Connection will be very PC like in terms of memory sharing and work sharing but that doesnt mean its going to be so much more powerful than its competition. When its talking about working as a companion its detailed right above. Memory sharing and Graphics process sharing just like the PC to GPU system of works. GPGPU is when the GPU is able to be programmed to do physics calculations on its own. you imght think it does but Nvidia has yet to come out with a statement that G70 or G72 will be able to do that after what ATI said recently about its processors.

 

[jvd]

I don't care what nobody else says from here on out. All I cared to read was this.

Notice that he says possible . NOt that they were done on the ps3 .

OPM: There is some speculation that many of the demos shown at E3 couldn't be done in real time on the PS3 hardware. Do you think this is true?
JH: There's no question the demos are possible on PS3 hardware.<<

Why would he say possible if they were real time ? He would have just said yes they were running on ps3 hardware .

 

[Shifty Geezer]

first its 20 GB write, 15 gb read, not 35 both ways as some might think.

Yes. I was saying there's a 35 GB/s BW total, which is bidirectional and not only one way. Hence data can be passed from RSX to Cell. There's obviously a reason they put that in there.

Physics Processing is done mostly through the Cell just like in Xbox 360 and RSX or Xenos in 1st gen games will hardly ever take part in any Physics detailing.

What 1st gen titles do and don't do with the hardware doesn't really have any bearing on the architecture though. You're arguing there's nothing new to RSX over G70. If there is something new but it's not used, it doesn't make it go away. eg. If MEMEXPORT isn't used on first-gen XB360 titles, you can't fairly argue that Xenos hasn't got a MEMEXPORT function just like a PC part hasn't got it!

Its true that the Connection will be very PC like in terms of memory sharing and work sharing but that doesnt mean its going to be so much more powerful than its competition. When its talking about working as a companion its detailed right above. Memory sharing and Graphics process sharing just like the PC to GPU system of works. GPGPU is when the GPU is able to be programmed to do physics calculations on its own. you imght think it does but Nvidia has yet to come out with a statement that G70 or G72 will be able to do that after what ATI said recently about its processors.

I wasn't saying RSX has got this functionality or that'll it'll ever do what Xenos does. We're talking about the atrchitectural differences between G70 and RSX, and you saying Jen-Hsun Huang has clarfied that RSX is nothing other than G70 with FlexIO working in the same way as XB's CPU and GPU. King Kenny has said that Cell and RSX can share data directly, such as RSX passing vertex data to Cell for physics processing. This suggests a closer relationship between CPU and GPU than you'll find (AFAIK) with a G70. Certainly a closer relationship than 'The EXACT same work process of Xbox 1.'

My point about the GPGPU was saying that if you look at a sentence and think about it, you can normally find at least two different ways to interpret it. English as a language (like many others) is full of ambiguities. As such you can't take a few lines ad lib'ed in an interview and conclude from that the entirety of the hardware's structure. No more so than KK saying RSX is a totally new chip or an nVidia dood saying RSX is derived from their PC part. All such statements are open to interpretation. Where you conclude RSX is a G70 part with upclocking and different memory interface (which is a point most people agree with in the main), I say there's nothing certain about the absence of other features not present in the G70 part. eg. Perhaps RSX features a MEMEXPORT like functionality which didn't come into Jen-Hsun Huang's statement? I'm not arguing that there is, for no-one's mentioned as much. Just that there's no proof of RSX architecture there, which you feel there is.

 

[j^aws]

...
As for whether RSX/Cell is coherant or not, I can't say but there is no requirement either way. There are 2 memory controllers in a PS3, one on Cell (essentially the EIB), the other on the GPU. This is the same as PC's today, when you access GPU RAM you have to mark the pages as non-cacheable. This is because the CPU can't snoop the read/writes, so can never know if the the stuff in cache is upto date. Now of course, its possible RSX interfaces so closer with the CPU mem controller that its knows the state of GPU RAM. Feel free to guess away, I can't comment until somebody can provide me with a public statement from Sony...

This bit I'm not clear on, with current PC's with a CPU and a TurboCache GPU, they are not cache-coherent, right? Is this because PCI-e does not have cache-coherency support in hardware, and the CPU-GPU bandwidth hasn't been great enough, right? FlexIO allows both cache-coherency and non cache-coherency in hardware. Now is this support flexible, i.e. where the developer has the option to choose, as there are probably circumstances where you'd prefer one over the other...?

 

[version]

http://patft.uspto.gov/netacgi/nph-...G&l=50&co1=AND&d=ptxt&s1=sony&OS=sony&RS=sony

 

[Shifty Geezer]

Drive by patent linking again. I don't think Version's latest hit is to do with PS3. Seem based on patents from 2001/02 talking about multipass rendering with single effects per pass. To my brief glance it looked like PS2 material.