Watch Impress PS3 Technical article, from GDC (some new info)

[PSman]

PSman its not a downgrade its more of a clarification that 1 of the 7SPE is assigned to run the OS, it was long rumoured that this would be the case. In reality it means that developers have 6 spe to play with.

Thank you for clearing that up:smile:

I doubt that, this is the RSX "full" specs (i could be wrong), though. Maybe Sony will revealed the RSX full spec at E3.

 

[Dave Baumann]

The fact that it states "NV47" based is more likely than not going to tell you the majority of the chip's story.

http://www.beyond3d.com/forum/showthread.php?t=29579

 

[DegustatoR]

Less ROPs might meen more complex ROPs. Although afaik RSX has the same functionality as any G7x...

 

[Titanio]

Sorry about the confusion with the VMX and memory latencies - I had become aware of those inaccuracies subsequently through GAF, but was unable to come back and edit the OP last night with all the down time. I'm glad the reporting didn't seem to confuse many, though

 

[Brimstone]

reguarding rsx,

does this give enough information to make more accurate guestimates on render/op throughput, eg, madd rate, filtered/unfiltered/combined texture rate, etc? assuming the pixel shader alu's in the rsx work in a similar fasion to G70?

I don't follow the inner workings too closly, but these specs would (to my maths at least) give a texturing advantage to xenos (when using all the units) and around about even shader performance (given XX% vertex usage)? or am I missing a whole lot? ;-) (I don't want to go making uneducated proclamations here )

It's all very facinating really.

The way I understand it, since Xenos is a unified shading architechture, the work load of vertex and pixel shading is balanced to make sure the execution units are always doing something. On a traditional split vertex and pixel shader GPU like the RSX, sometimes pipelines are sitting idle waiting to do some work, which obviously prevents it achieving anywhere near peak performance.

 

[Jadoku]

The way I understand it, since Xenos is a unified shading architechture, the work load of vertex and pixel shading is balanced to make sure the execution units are always doing something. On a traditional split vertex and pixel shader GPU like the RSX, sometimes pipelines are sitting idle waiting to do some work, which obviously prevents it achieving anywhere near peak performance.

The way I understand it a unified shading architechture is no advantage at all in a closed system. But very usefull in a computer running different games optimized for no special configuration. I don´t know much about this, if there really is a benefit from unified architechture please explain why.

 

[nAo]

The way I understand it a unified shading architechture is no advantage at all in a closed system.

Certainly it's not as useful as in an open system, but it can be a relevant advantage, especially in the early days of a console when one still does not have a clear idea about the hw and how to take advantage of it.

Marco

 

[joebloggs]

Nao can you expand on RSX details any more than the vague information currently floating around the ether*.





* I will protect you from the Sony Ninjasâ„¢

 

[Rockster]

It's certainly an advantage, because the case you described as ("closed box") is the developer modifying the application to suit the hardware, as opposed to a unified shader architecture where the hardware modifies resources to suit the application. I can think of many instances where it would be of benefit to have the ability to increase vertex shading vs. pixel shading and vice-versa, and not be forced into always using a consistent ratio. The other benefit to the design is it should be better suited to mask latency and better handle branching.

 

[nAo]

better handle branching.

Well..it depends, a not unified architecture as G70 is way better than Xenos at dynamic vertex branching.

 

[pipo]

...especially in the early days of a console...

You mean like, the first couple of years?

 

[Rockster]

Well..it depends, a not unified architecture as G70 is way better than Xenos at dynamic vertex branching.

I can't think of an instance where that would be the case. It's obvious to me that G70 is way better at vertex branching than it is at branching within a pixel shader, but how are you translating that to Xenos. Can you elaborate?

 

[Platon]

I can think of many instances where it would be of benefit to have the ability to increase vertex shading vs. pixel shading and vice-versa, and not be forced into always using a consistent ratio.

That was a bit what I was thinking as well, having the unified architecture I can see that it enables developers a bit more freedom in how they make their games...

 

[Zeross]

I can't think of an instance where that would be the case. It's obvious to me that G70 is way better at vertex branching than it is at branching within a pixel shader, but how are you translating that to Xenos. Can you elaborate?

Xenos ALUs are working on vectors of 64 elements, it means that dynamic branching is only a win if all 64 elements take the same branch.

G70 vertex shader on the other hand are MIMD, the performance penalty when two vertices takes divergent branches is close to zero.

 

[Carl B]

Well, I have to say there's a sense of whiplash to come back and learn that the Cell's PPE is in fact the same as before. There was a slide and everything! Oh well oh well. Still I hope the 'secrets' of DD3.1 are revealed to us one day, because I can't buy into the yield theory alone on that one; and yes granted that's just based on my feeling a yield-related revision would warrant a 0.x step vs a x.0 step on the revision code.

Anyway I'm not expecting any answers on this one per se, but does the abundance of compute power an SPE represents vs background code run on the PPE indicate some plan Sony has for... well and I guess this is the blank I'm looking to have filled. Kutaragi's persistent 'Matrix' Cell-grid world? DVR/LocationFree/Other functionality preserved while in game? Maybe any or all of these, maybe just set aside for possible future ideas/uses - but is there a sense as to what sort of applications would lead SCE to make that call? It seems overkill for the normal case system app, which is what has me thinking there must be something beyond the normal case useage scenario planned on the system side.

 

[Titanio]

Anyway I'm not expecting any answers on this one per se, but does the abundance of compute power an SPE represents vs background code run on the PPE indicate some plan Sony has for... well and I guess this is the blank I'm looking to have filled. Kutaragi's persistent 'Matrix' Cell-grid world? DVR/LocationFree/Other functionality preserved while in game?

It seems they want to preserve most OS functionality while in-game.

I would hazard a guess that the PPE time taken is quite minimal, if an SPE is being reserved - I'm sure they'll offload as much as is physically possible to that SPE to get their money's worth, so to speak.

 

[Carl B]

It seems they want to preserve most OS functionality while in-game.

I would hazard a guess that the PPE time taken is quite minimal, if an SPE is being reserved - I'm sure they'll offload as much as is physically possible to that SPE to get their money's worth, so to speak.

I completely agree, I just wonder as to what this functionality would be in the first place to warrant tying up a significant amount of one's processing power. I would think that there must at least be something in the pipes to warrant the SPEs 'power,' though of course I could just as easily envision the situation as being 'preserve PPE cycles at all costs.'

 

[liolio]

can it be related to a software layer that could help Code created with tools like rapidmind to run in more effective manner?
+ os and network helper threads.

 

[expletive]

It seems they want to preserve most OS functionality while in-game.

I would hazard a guess that the PPE time taken is quite minimal, if an SPE is being reserved - I'm sure they'll offload as much as is physically possible to that SPE to get their money's worth, so to speak.

It'd be interesting to see what work the OS takes care of. Reserving 1 SPE for OS doesn't necessarely have to be a bad thing if other processes that are going to be used by games anyway are handled by the OS by default. If they can pack as much functionality into the OS, it could take away some work for developers and give some default functions that practically every game will use because "it's there to be used and on anyway".

Another thing that may require an entire SPE that would be "OS related" is video scaling. I've been wondering how the PS3 will handle scaling of the natively rendered frame to various output resolutions. I imagine that even if a game isnt rendering at 1080p, Sony will still offer it as an output resolution (if only for marketing), as well as 1080i, 13xx x 768, etc. They could use a dedicated IC but that would seem contradictory to what the cell 'ideology' for CE devices is. My guess is that would be most cost effective to use a portion of the Cell for this, and scaling 720p to 1080p is no small feat.


Can someone explain what the PPE is being used for? Is it essentially 'another SPE' or reseerved for something else?

 

[Titanio]

Another thing that may require an entire SPE that would be "OS related" is video scaling. I've been wondering how the PS3 will handle scaling of the natively rendered frame to various output resolutions. I imagine that even if a game isnt rendering at 1080p, Sony will still offer it as an output resolution (if only for marketing), as well as 1080i, 13xx x 768, etc. They could use a dedicated IC but that would seem contradictory to what the cell 'ideology' for CE devices is. My guess is that would be most cost effective to use a portion of the Cell for this and sclaing 720p to 1080p is no small feat.

Ja, that would definitely be one ever-present requirement, although it wouldn't surprise me totally if that was something RSX might handle (?). I think more generally they're serious about have PS3 as an "always/on connected" server and media player, even if someone is playing a game at the same time.

Can someone explain what the PPE is being used for? Is it essentially 'another SPE' or reseerved for something else?

The SPE in and of itself cannot run an OS. You need some code on the PPE to control things, but I'd say most functionality beyond what is barely required will be on the SPE.

It'll be interesting to see how difficult hackers find it to compromise PS3's OS also. I'm sure that reserved SPE will be fenced off on a hardware level as described in those early Cell/security whitepapers. I wonder if that will make things any more difficult for pirates and the like.