RSX = Stream Processor!?!

leechan25 said:
So you don't like the way I said something...

I was not looking to start a fight but I'll glad to finish one. You may not remember but you and I were talking back then about the same topic. look it up. So I was continuing what was talked about. I don't like your tone. Sounds like my wife when she can't get her way. If you can't talk about things you disagree without talking down to people maybe you should not comment.

Feel free to finish the 'fight' anytime.
 
Jaws said:
GPUs are ALREADY multicore. They have multiple fragment quad processors and multiple vertex processors. And they are ALL stream procerssors.


I guess i'm thinking of it being more like cell SPE's.
 
xbdestroya said:
Rog but you have to tell us what we're supposed to derive from that. ;)

That's from July 22nd and it was discussed heavily here back in the day, as I'm sure you know. I know your theory of PS3 involves heavy cooperation between Cell and RSX, possibly with the dedicated 'slaving' of SPE's, but no one's disagreeing with you. In fact Cell:RSX communication and assistance has been a central point of what most of us expect from the PS3. It's simply given that theory, there's too little known to either support it or shoot it down. Me personally, I think it sounds viable.

Great post btw Shifty, I'd rep ya but I gotta 'spread more around' first.

I'd love to make more of the RSX than that...but I think that's really all it is.

But if you think for a second that programmable DMA logic may have been the missing link in what will make the RSX work so well with the cell, as well as the Flex I/O...and the physical make-up of the CELL itself starts making more sense in the realm of the PS3 system as a whole. In fact they may reuse the PS2 GS (or was it EE?) with the integrated 4 MB VRAM Cache with some alterations as the programmable DMA logic I speak of just like they used the ps1 chipset in the ps2 for I/O and bc. How about them apples Nerve-Damage? ;)

In fact, it's really the most viable architectural ideology we can extrapolate from what Sony has presented to us because the RSX can access both pools of memory and has massive bi-directional bandwidth between itself and CELL. What is the RSX using that inter-chip bandwidth for...sharing/slaving processing power and memory access to reduce data redundancy.
 
leechan25 said:
I guess i'm thinking of it being more like cell SPE's.
Why? What advantage would that provide over the existing fragment/vertex processor system? (or over xenos' array of unified shader elements, for that matter?)

One of the reasons people are questioning your comments is that you appear to be throwing around buzzwords with little thought put into the logistics or consequences of applying them to the system being discussed. You seem to be thinking that the RSX will be some amazing, innovative, highly-custom design with features we've never seen before, based on little other than faith.
 
I can't believe how hostile you guys are getting over this. Even between people who are on the same side.

Like you guys I want the best for the PS3. Let's tone down our opinions till the facts are released.
 
Jaws said:
And why should they be SPUs?

Sony decide Cell need to be a multi-core CPU in order execute tasks simultaneously and independently. I think the GPU partnership would be stronger if its cores paired with Cell SPE's in order complete its task and balancing the processor workload. What good would RSX be Cell if the CPU overloads GPU with instructions slowing down the Cell killing the potential computing throughput. SPE's will be sending instructions for physics processing, ray casting, and whatever esle cell can throw at the GPU simultaneously. How is it going to handle all of that simultaneously and independently if needed? I think with it own pair of processor cores design to allow the GPU rotation task between it and the Cell. Sony want PS3 to do GI, Raycasting, and more in realtime. These task eat up so much computing power that it would foolish to trying because the architecture can't support it even with PS3 wide data bus and fast memory. However, if the RSX could pair one of its cores with an SPE to only work on say Raytracing independently of whatever else the hardware is doing it would not slow up or prevent of task from preforming. Everything could be balancing out to incease throughput.

That's what I think.
 
leechan25 said:
Sony decide Cell need to be a multi-core CPU in order execute tasks simultaneously and independently. I think the GPU partnership would be stronger if its cores paired with Cell SPE's in order complete its task and balancing the processor workload. What good would RSX be Cell if the CPU overloads GPU with instructions slowing down the Cell killing the potential computing throughput. SPE's will be sending instructions for physics processing, ray casting, and whatever esle cell can throw at the GPU simultaneously. How is it going to handle all of that simultaneously and independently if needed? I think with it own pair of processor cores design to allow the GPU rotation task between it and the Cell. Sony want PS3 to do GI, Raycasting, and more in realtime. These task eat up so much computing power that it would foolish to trying because the architecture can't support it even with PS3 wide data bus and fast memory. However, if the RSX could pair one of its cores with an SPE to only work on say Raytracing independently of whatever else the hardware is doing it would not slow up or prevent of task from preforming. Everything could be balancing out to incease throughput.

That's what I think.

Let me get this clear, you think the more SPUs are needed, and that these SPUs should be in RSX?
 
Edge said:
I can't believe how hostile you guys are getting over this. Even between people who are on the same side.
I'm only seeing one guy being hostile. Maybe some people are reading confrontation into posts where none is meant? It's not hard to read an honest question as a rhetorical putdown at times.
 
Jaws said:
Let me get this clear, you think the more SPUs are needed, and that these SPUs should be in RSX?

No it does not have to be a SPE, just like cell's but the RSX should have some kinda multicore setup for working "better" with cell simultaneous and independent SPE's.
 
Shifty Geezer said:
I'm only seeing one guy being hostile. Maybe some people are reading confrontation into posts where none is meant? It's not hard to read an honest question as a rhetorical putdown at times.

You think this is an honest question:

"So now in your own words, with the massive evidence you have at your disposal, please tell us exactly what the RSX is, since apparently it is clear as day to you?"

or is it a rhetorical putdown?
 
leechan25 said:
No it does not have to be a SPE, just like cell's but the RSX should have some kinda multicore setup for working "better" with cell simultaneous and independent SPE's.

Okay, that could be something as simple as what XeCPU<->Xenos do with L2 cache sharing. Remember RSX is connected to FlexIO, which is connected to the EIB, so there could be some interesting possibilities there with CELL<->RSX and cache coherency with the PPE L2 etc... and like I said, GPUs are already multicore...
 
I don't care if the RSX turned out to be a paper airplane.

I just want to know now dammit!

Damn you Sony for keeping the techies (AKA extreme minority population of even early adopters) in the dark.
 
leechan25 said:
No it does not have to be a SPE, just like cell's but the RSX should have some kinda multicore setup for working "better" with cell simultaneous and independent SPE's.
Why?
 
Jaws said:
Okay, that could be something as simple as what XeCPU<->Xenos do with L2 cache sharing. Remember RSX is connected to FlexIO, which is connected to the EIB, so there could be some interesting possibilities there with CELL<->RSX and cache coherency with the PPE L2 etc... and like I said, GPUs are already multicore...

Okay, but can these cores work simultaneously and independently on task like Cell's SPE's?
 
leechan25 said:
You think this is an honest question:

"So now in your own words, with the massive evidence you have at your disposal, please tell us exactly what the RSX is, since apparently it is clear as day to you?"

or is it a rhetorical putdown?

Yeah I admit, that was a rhetorical taunt. You seemed overly defensive and hostile about your RSX architecture, and frankly I wanted to hear your vision for the RSX since you seem to reject the G7x evolution one out of hand (while tossing many an insult at the same time).

leechan25 said:
No it does not have to be a SPE, just like cell's but the RSX should have some kinda multicore setup for working "better" with cell simultaneous and independent SPE's.

As far as this goes, I think you do not understand the nature of GPU's and how they process data. Why add SPE's to 'reduce the burden' on RSX when you could just add more quads instead? Do you not agree that these quads represent inherently parallel 'cores' and structures? When one looks at the die area of a core and determines they can 'add more,' I think the question a lot of us are wondering with regard to the SPE crowd, is why they (you leechan) feel an SPE would be the best use of die area. Because 'more cores' is not a valid answer; three pixel quads could be considered as much to be three cores as three SPE's on Cell could be considered to be three cores, so why is the addition of more of the units GPU's naturally contain not seen as a valid route?
 
leechan25 said:
Okay, but can these cores work simultaneously and independently on task like Cell's SPE's?

Given G70, there are 8 MIMD vertex processors and 6 SIMD fragment quad processors that can work independently on separate threads. These processors are also fully programmable via shaders, e.g. Cg... and also have a more graphics capable ISA compared to SPUs...
 
leechan25 said:
You think this is an honest question:

"So now in your own words, with the massive evidence you have at your disposal, please tell us exactly what the RSX is, since apparently it is clear as day to you?"

or is it a rhetorical putdown?

I'd say it was a pretty valid response/question to your comment about the "smarty pants" crowd now desperately trying to invalidate the patents. (which seemed curiously funny to me, as the recent information we've gained only seem to further invalidate the patents)

When reality stares you in the face, don't look away. Trying to take information from patents filed years back (or hell, even recent patents) and base your views of something on it, in spite of logic (and functionality), will often get you some odd looks. The need for SPEs in the RSX is non existant (or any other elaborate additions). It doesn't make sense to me why SPEs would need to be there in the first place... why not just hope for 32/10 shaders or something (as, while that's still wishfull thinking, it's at least plausable and possibly useful)?

I think you're getting some (possibly unwarranted) sarcasm because the idea that some of you are proposing is so unnecessary and almost absurd that is reeks of the "More is better, right? SPEs are good, right? lets add SPEs everywhere!" kind of thinking. I had to bite my tongue a few times in responses to this stuff...
 
Last edited by a moderator:
xbdestroya said:
Yeah I admit, that was a rhetorical taunt. You seemed overly defensive and hostile about your RSX architecture, and frankly I wanted to hear your vision for the RSX since you seem to reject the G7x evolution one out of hand (while tossing many an insult at the same time).



As far as this goes, I think you do not understand the nature of GPU's and how they process data. Why add SPE's to 'reduce the burden' on RSX when you could just add more quads instead? Do you not agree that these quads represent inherently parallel 'cores' and structures? When one looks at the die area of a core and determines they can 'add more,' I think the question a lot of us are wondering with regard to the SPE crowd, is why they (you leechan) feel an SPE would be the best use of die area. Because 'more cores' is not a valid answer; three pixel quads could be considered as much to be three cores as three SPE's on Cell could be considered to be three cores, so why is the addition of more of the units GPU's naturally contain not seen as a valid route?

I think a multicore GPU is needed because the task because asking of this GPU will be changed from traditional GPU. There never been a GPU that could preform GI, RT, and other task done by a farm of CPU's. Sony is asking this of it GPU and Cell. I mean look at the PPU, gpu's could handle it tasks with no problems before. However, the standard has change and the needs are greater so hardware designers created the PPU. more is needed from the GPU that's going to create realtime CG games that look like movies.
 
Bobbler said:
I'd say it was a pretty valid response/question to your comment about the "smarty pants" crowd now desperately trying to invalidate the patents. (which seemed curiously funny to me, as the recent information we've gained only seem to further invalidate the patents)

When reality stares you in the face, don't look away. Trying to take information from patents filed years back (or hell, even recent patents) and base your views of something on it, in spite of logic (and functionality), will often get you some odd looks. The need for SPEs in the RSX is non existant (or any other elaborate additions). It doesn't make sense to me why SPEs would need to be there in the first place... why not just hope for 32/10 shaders or something (as, while that's still wishfull thinking, it's at least plausable and possibly useful)?

I think you're getting some (possibly unwarranted) sarcasm because the idea that some of you are proposing is so unnecessary and almost absurd that is reeks of the "More is better, right? SPEs are good, right? lets add SPEs everywhere!" kind of thinking. I had to bite my tongue a few times in responses to this stuff...

Okay Whatever,
 
Back
Top