PS3 Cell, RSX, SDK, OS, Linux INFO

Where does it say RSX has more shader power??? It does say that X360 devs are having a hard time getting shader performance currently, but that it will change once they get more familiar with USA.

It was speculated for ages here that RSX has more shader power. This quote pretty much confirms it:

Says developers are having exact opposite problems with the PS3 compared with the 360, where they aren't having no memory bandwidth problems due to the 10BM eDRAM but have shader performance issues compared to the PS3. Saying they are lacking in Shader ALU performance and having stalls due to not having enough "threading resources".

Source of translation

Anyway, XBD also agreed it's the case and he's usually pretty informed on these matters due to his dev connections.
 
There's a difference between system OS and the user operating system. It's like Amiga's OS was on it's kickstart ROMs, but the OS people used, AmigaDOS and Workbench, was loaded from disc. PS3OS system OS is in Flash, Linux or other user OS's is on HDD, and Linux (supposedly, still not confirmed!) comes preinstalled on the HDD.

So in your words the system os would be "bios" and the user os would be Windows in a regular PC ?.
 
Kind of, only the BIOS is very extensive. It'd be like the PSP's OS, on Flash and with plenty of functions. It'd be upgradeable too. Linux on the PS3 wouldn't be the OS, but would be an application.
 
Anyway, XBD also agreed it's the case and he's usually pretty informed on these matters due to his dev connections.

Ok now let's step back a minute here... :) ;)

It's not anything to do with speculation or dev connections or anything else - it's just that the RSX does have more shading 'power,' based on the specs of each chip. Now, I also pointed out and backed Acert up though when he was highlighting that even though RSX has more 'power,' RSX is not - featurewise - as well equipped to handle some of the future directions shader instructions might go in.

So... although I think it's fair to say and highlight that RSX has more shader power, at the same time I think we should always add the caveat that Xenos has the more robust support in terms of 'Shader Model' (to put it in DX terms) evolution.

What, if anything, that will eventuallly mean for graphics on Xenos will simply be up to the effort we see developers put into things such as branched shaders as time goes on.
 
Last edited by a moderator:
Xenos also has the capability for much more vertex shading correct?

And shouldn't the efficiency of Xenos increase as developers get used to working with, and exploiting unified shader arrays? I know in raw numbers RSX has an edge, but the USA's claim to fame has always been increased efficiency. Right?
 
Xenos also has the capability for much more vertex shading correct?

Yeah, for sure. Now granted, more on the vertex means less on the pixel shading... so it's something important to keep in mind, but Xenos definitely has the higher vertex throughput potential.

I want to add that there are some strong indications that Cell will play an ever increasing role with what the PS3 does with it's vertices, so it's not like vertex work is a cut and dry comparison either, but yeah in terms of throughput, Xenos.

And shouldn't the efficiency of Xenos increase as developers get used to working with, and exploiting unified shader arrays? I know in raw numbers RSX has an edge, but the USA's claim to fame has always been increased efficiency. Right?

I don't know about 'exploiting' - personally I thought that USA was supposed to be completely transparent to devs working with it, but if there's some other thing afoot, I just haven't kept up with it. I'm sure someone that has will likely post on the matter. In terms of distributing workloads among the ALUs however - and thus hardware utilization - Xenos shoudl enjoy an efficiency edge though.
 
I'm not quite sure I really understand that argument... isn't the whole point of unified shading to alleviate any issues with having to match a specified ratio of VS/PS? There shouldn't be any inherent issues with trying to max out a unified architecture, it does that itself -- that's the whole point of it, it has an inherent flexibility so developers could throw whatever they wanted at it within a certain max amount of ALU performance (between VS and PS). If they are having ALU performance issues, that isn't likely to be the kind of thing that will drastically change (beyond the average change in programming competence that comes over the lifetime of a platform -- which is seemingly quite a bit). If they are having issues with USA, then they'd have equal or possibly more issues with a non-unified architecture; the same improvements would apply to both.

I'm not trying to say RSX is more powerful than Xenos or anything, I just don't buy the "leaning the USA" as any sort of special case that only applies to Xenos. I think both GPUs are pretty close to each other (different bottlenecks at any given point, but definitely same year parts).
I agree with Bobbler that Xenos and RSX are "same year" parts and on average their performance should be similar with each having strengths in certain areas. I disagree with people claiming RSX has more shader power than Xenos and I disagree that the quote saying a lack of "threading resources" proves it. If threading resources are an issue the problem is likely not having enough threads to cover latency and ALUs might even be idle. In the end the chips are close enough that comparing them based on specs is a lost cause. Of course this applies equally to most PC chips.

That's a little off track from Bobbler's topic though. A unified shader architecture should just work and load balance properly, but that doesn't mean devs can be naive when they write shaders. Xenos is very different from previous ATI chips and likely needs different optimizations to get the most out of it. The Gamefest presentations highlighted this well. One presentation talked about packing data to minimize vfetches and how poor packing can nearly double the latency of a vertex shader.
 
I don't know about 'exploiting' - personally I thought that USA was supposed to be completely transparent to devs working with it, but if there's some other thing afoot, I just haven't kept up with it. I'm sure someone that has will likely post on the matter. In terms of distributing workloads among the ALUs however - and thus hardware utilization - Xenos shoudl enjoy an efficiency edge though.

Surely changes in the rendering engine, or the way developers approach the games technical design could exploit the more flexible USA, no? It seems to me that, although the USA's are fully automated, it will still require some changes to convential thinking to take full advantage of the possibilities it provides.
 
Back
Top