New dev tools ? PlayStation 3 Edge

[nAo]

The fundamental problem to the argument. All we can ever have, thanks to NDAs, is different sides expression a truth without an explanation of the truth.

Yes..and no. cause it seems to me that someone here has explicitely stated he does not know what's wrong with it, not that he can't talk about it due to NDAs.
NDAs don't really matter here, if you say something suck you should also understand why, otherwise as a developer you really did a poor job/exercise imho.

 

[Brimstone]

Yes..and no. cause it seems to me that someone here has explicitely stated he does not know what's wrong with it, not that he can't talk about it due to NDAs.
NDAs don't really matter here, if you say something suck you should also understand why, otherwise as a developer you really did a poor job/exercise imho.

But it is developers "plural" that have been mentioned. The Japanese article echoed the sentiments of Joker, by saying the RSX vertex shaders are an area of concern for many PS3 developers.

 

[nAo]

At least I can now say that I know how urban legends are born

 

[joker454]

Yes..and no. cause it seems to me that someone here has explicitely stated he does not know what's wrong with it, not that he can't talk about it due to NDAs.
NDAs don't really matter here, if you say something suck you should also understand why, otherwise as a developer you really did a poor job/exercise imho.

I've already stated two of the reasons, both of which were dismissed without any reason other than "they are wrong". The first being that you pay a penalty for every vertex input past the second one, the other being that xenos has *far* more resources allocated to vertex processing. These two give it an edge in vertex processing. Now, since those two reasons aren't "accepted" we need to look for other reasons. Beyond the above two it's true, I don't really know what else gives is such an edge. My friend at work will know, given that he knows the designer of the rsx, and given his extreme experience in that field. There's always someone smarter across the hall and in this case I'd have to defer to his knowledge over mine.

I don't understand why it's hard to accept that one video card is better than another. Historically, ATI and NVidia have gone back and forth with each holding the performance crown at different times over the years. It's just isn't all that surprising to me that in this case an NVidia 7 series card will lose, especially considering its actually a relatively old architechture (which has since been dropped in the 8 series).

 

[inefficient]

...

Did you write the 360's VS shader code as well? Or is your project shoe horning an existing project to the PS3? Because it sounds like the later to me.

 

[chachi]

Probably some will think I sound boring or fanboish but to be honest I don't care I stand with my opinion on the subject ppl can write even 10k times that RSX sucks at VS (and it seems to me they don't even know why..) but they're wrong. Maybe they think the problem is in VS, but this doesn't automatically make this true.

To be honest I think to most objective observers you probably do come across as a bit of a Sony fanboi, but there's not really anything wrong with that. It's nice to see developers enthusiastic about the platform they're working on, it certainly beats the alternative. The only problem anyone should have with "fanboi-ism" is when someone is trying to be disingenuous about it, proclaiming their objectivity when in reality they don't have any. OK, trolling is pretty bad too, but that's why you have a killfile.

Back on point...

For my own experience HS is the vast majority of times not VS limited and I think we push quite a lot of geometry on screen, both with simple (zpre pass and shadow maps) and complex rendering passes.

The problem here isn't necessarily that RSX sucks at VS or that it can't do a lot of geometry, it's that compared to the Xenos there's a growing consensus that it doesn't seem to be as capable. If it weren't an issue would Sony bother having SPUs do culling?

If you're not writing a multi-platform game then it doesn't matter, you won't have any problems at all, and from what I've seen most of Sony's big budget games (including HS, which from what I saw on Heroes is frickin' amazing and all the NT guys should be more than proud of their work) look fantastic. If you're are working xplat though then things get more interesting because you don't have that budget and you do have one platform that is busting your balls more than the other one. Time spent fighting with one platform to get it to display the scene the same as the other one is less time that you could have spent improving something that would take advantage of that platform's strength. This is also time you could have spent improving the "strong" platform so it's not like it's really doing that one any favors either.

As people become more familiar with both and build up a code library this won't be as big of an issue, though it'll probably swing the other way with processor expensive things like physics (though that's not as obvious an issue IMO as you don't get screenshots of physics and people around the world screaming about every minor nitpicky detail like you do with graphics.)

 

[inefficient]

The problem here isn't necessarily that RSX sucks at VS or that it can't do a lot of geometry, it's that compared to the Xenos there's a growing consensus that it doesn't seem to be as capable. If it weren't an issue would Sony bother having SPUs do culling?

Not even nAo is arguing that the Xenos is not more capable than the RSX with regards to vertex shaders. His point is that what is available is more than sufficient for handling the types of task jokers is complaining about having problems with.

Xenos has a significant advantage in that it can allocate far more resources just for VS than the RSX ever could. But the point of that advantage is rather moot when in the vast majority of cases those resources are going to be tied up doing pixel shading (as SMM explained) But of course if your game is not too concerned about pixel quality, then great, now you can throw even unoptimized code into the Xenos and still get decent results.

Some people keep bringing up that to really push the PS3 you have to make the trade off of using the SPU power. But Xenos' VS power comes at the expense of PS power in just the same way. Everything about game development is about trade offs. Don't look at trade offs as something bad. Look at them as opportunities.

 

[joker454]

Not even nAo is arguing that the Xenos is not more capable than the RSX with regards to vertex shaders. His point is that what is available is more than sufficient for handling the types of task jokers is complaining about having problems with.

Hmm, well I didn't read his comments that way. If I was wrong, then my bad. To me it sounded like he was saying that the statement of "rsx has vertex issues" was plain wrong. If he did indeed mean it the way you're saying above, then I mostly I agree. I say "mostly" because there is still a tens of megabytes of free memory difference between the two machines, so the 360 version will still have an edge there for our title. This memory difference can be very tricky to overcome depending on your title. I also say mostly because culling only gets you so far. What if you actually have to render a few million visible triangles? If your scene has hundreds of humanoid type meshes then backface culling on the spu's will pay off big time since half the geometry will never make it to rsx. But what if you need to show architectural type geometry thats laid out such that you can only cull 10% of the geometry on the spu? Inevitably that means the rest of it still needs to be fed to the gpu, which means you are still dependent on rsx vertex performance.

 

[patsu]

Hmm, well I didn't read his comments that way. If I was wrong, then my bad. To me it sounded like he was saying that the statement of "rsx has vertex issues" was plain wrong. If he did indeed mean it the way you're saying above, then I mostly I agree. I say "mostly" because there is still a tens of megabytes of free memory difference between the two machines, so the 360 version will still have an edge there for our title. This memory difference can be very tricky to overcome depending on your title. I also say mostly because culling only gets you so far. What if you actually have to render a few million visible triangles? If your scene has hundreds of humanoid type meshes then backface culling on the spu's will pay off big time since half the geometry will never make it to rsx. But what if you need to show architectural type geometry thats laid out such that you can only cull 10% of the geometry on the spu? Inevitably that means the rest of it still needs to be fed to the gpu, which means you are still dependent on rsx vertex performance.

Perhaps it's the context...

Did I understand you correctly if I say the following (Note that I'm not a graphics person):

(A) RSX's vertex limitations is _only_ in vertex setup.

(B) The RSX vertex processing power is sufficient to your title's workload

(C) SPU Culling will help (A) and (B), but mostly (A)

(D) If one needs to render millions of visible triangles beyond RSX's vertex setup limits, will you run into some other bottleneck first (or after) even on Xenos for a high quality output (e.g., 720p, 4xAA, at least 30fps) ?

(E) If one needs to render millions of visible triangles, are there suitable techniques (besides culling) to reduce the triangles -- say via the SPUs -- for similar visible output quality ?

(F) As for the memory arguments (it's really a separate point), are there techniques to overcome it ? It seems that your title is tailored for Xbox 360. Perhaps if it's coded for RSX, the way graphic objects are created, recreated, or stored would be optimized for RSX.

I'm asking/validating these questions because high quality output seems possible on PS3. May be it's more straightforward to implement them on Xbox 360 (because the sub-systems are self-contained). In PS3, we'll need to spread the load between the CPU and GPU by design. If one were to start his design first with PS3, is there any (other) advantage for the flexible integration between CPU and GPU ?

 

[Carl B]

Guys I think Joker needs to be shown a little more credit than he's being given here, y'know?

He needn't be viewed by some as an adversary or as selling RSX short. Everyone developing for these systems has a legitimate angle from whence they approach things, and we all learn from the discussion. Even were someone to be out-and-out right, and someone wrong, there's shades of grey in there that are important to the larger understanding of what choices are made in what situations in a multi-platform world.

 

[patsu]

Oh dear... do I sound rude ? I was thinking out loud... because it seems that both of them (views represented by nAo and joker454) can be correct under different contexts.

 

[Carl B]

Patsu it wasn't directed at you, believe me. Or anyone in particular per se either. It'd just be nice if Joker's posts could be addressed on message alone and discussed as such rather than with a sort of underlying sense that some view him as on a mission to discredit RSX or something. And again, that's not specific to you or anyone, but a sort of subtle undertone in the thread.

 

[archangelmorph]

Patsu it wasn't directed at you, believe me. Or anyone in particular per se either. It'd just be nice if Joker's posts could be addressed on message alone and discussed as such rather than with a sort of underlying sense that some view him as on a mission to discredit RSX or something. And again, that's not specific to you or anyone, but a sort of subtle undertone in the thread.

Well its kind of hard when he uses language like "RSX sucks" when what he claims he is really saying is "RSX only such when compared with the mighty Xenos.." which after reading every thread in this post so far, I don't think anyone is disputing..

I think if you're going to downplay the performance of one platform component over the other then you still need to be careful about how you put it across (since saying something "sucks" makes it very difficult to be intepreted in a "relative" context..)

That is where I believe the heart and soul of this arguement lies and why, after Joker cleared things up with his last post, we should probably get back to talking about edge...

 

[Shifty Geezer]

So we have a reason now! It's the issuing of vertex attributes where RSX is at a disadvantage (actually this info appeared in the original Joker454 thread, but it didn't stick ). This post from SMM explains the problem nicely.

Is this because of Xenos V5 vectors versus RSX's V4? That 5th vector dimension sees a lot of use, and RSX needs to double up fetches/issues to the same info?

 

[one]

Going back to Edge, it's interesting to see how GCM Replay can contribute to RSX graphics tuning. Probably those who complain about RSX were doing their work in a very poor debugging environment.
http://www.watch.impress.co.jp/game/docs/20070316/pe39.htm
http://www.watch.impress.co.jp/game/docs/20070316/pe40.htm

 

[Panajev2001a]

I've already stated two of the reasons, both of which were dismissed without any reason other than "they are wrong". The first being that you pay a penalty for every vertex input past the second one, the other being that xenos has *far* more resources allocated to vertex processing. These two give it an edge in vertex processing. Now, since those two reasons aren't "accepted" we need to look for other reasons. Beyond the above two it's true, I don't really know what else gives is such an edge. My friend at work will know, given that he knows the designer of the rsx, and given his extreme experience in that field. There's always someone smarter across the hall and in this case I'd have to defer to his knowledge over mine.

I don't understand why it's hard to accept that one video card is better than another. Historically, ATI and NVidia have gone back and forth with each holding the performance crown at different times over the years. It's just isn't all that surprising to me that in this case an NVidia 7 series card will lose, especially considering its actually a relatively old architechture (which has since been dropped in the 8 series).

Do you work at IOI by any chance and know a guy that goes by the nick-name of sparky ?

 

[nAo]

The problem here isn't necessarily that RSX sucks at VS or that it can't do a lot of geometry, it's that compared to the Xenos there's a growing consensus that it doesn't seem to be as capable. If it weren't an issue would Sony bother having SPUs do culling?

I never wrote a thing about 360 vs PS3 so ppl will not draw me into this argument. Regarding what Sony provides to devs..see that as an answer to a question, but the question is not the one being asked in this thread over and over again. If I keep saying VS are ok there must be a reason..

 

[nAo]

I've already stated two of the reasons, both of which were dismissed without any reason other than "they are wrong".

They were not dismissed, they were properly addressed since it's something really easy to do for anyone, multiplatform devs included.

These two give it an edge in vertex processing. Now, since those two reasons aren't "accepted" we need to look for other reasons.

The problem is that you think those things give a platform an edge pver the other one: I say that these things are a no brainer and if there's a problem is somewhere else. You speak like you got all your facts and you know the platform very well.
I have no doubt that in a year or so, when you will have more work experience you will have changed your mind. Maybe you wil still see some problems, but you will see them somewhere else..

I don't understand why it's hard to accept that one video card is better than another.

I never said a word about that, I've just addressed your comments about RSX.

 

[V-G]

nobody is saying it's a POS.

I tend to think that both systems are close to EQUAL (too close to call- see my sig) or at least will be, as Cell gets leveraged with tools like EDGE.

I'm not following anything blindly either, I've been reading about both of these architectures for 2 years with an open mind. It's common knowledge that Xenos is more efficient than RSX and that Cell needs to help RSX in some capacities. this is just confirmation from someone running both.

If it ruffles some feathers here to come to the conclusion that the subsequent strengths and weaknesses of each architecture could/would allow them to perform nearly equally in real world scenarios, then I'm sorry they feel that way.

I know little to nothing about graphics prossing so it would be futile of me even to go there.

But with RSX being equivelent to a 7900 GPU for PCs where there's on CPU to help out with VS it would surprise me if the vertex shaders have been trimmed down unless the transistor budget went somewhere else (with VGA engine removed, DirectX instruction set removed, RSX having a bigger die size and so on I can't imagine it even though some of the budget has been used for redundancy).

In the end it's all up to the devs and I won't really care about specs or how hard a platform is to develop for as long as it can't be seen in the games.

It's just isn't all that surprising to me that in this case an NVidia 7 series card will lose, especially considering its actually a relatively old architechture (which has since been dropped in the 8 series).

That's a weird comment, didn't expect to hear that kind of reasoning from a dev.
It sounds like something I could have written in my post above.

Now that were are in this stage of debate, to my knowledge unified as in flexible costs more transistors than specialized pipelines compared to the throughput, isn't that true? The way I understand it is that the unified approach is more efficient on open platforms like PCs where the code can't be balanced after the amount of VS and PS pipelines in each GPU, but in a closed system the performance increase isn't as noticable as the code can be tailored for the specialized pipelines (unified will still have an advantage in ease to develop for but not performance).

Please help me out as I know nothing about the math behind the rendering and for that reason do not know how many elements VS and PS have in common.

Oh dear... do I sound rude ? I was thinking out loud... because it seems that both of them (views represented by nAo and joker454) can be correct under different contexts.

Well please keep sharing.

Patsu it wasn't directed at you, believe me.

Jedie mind trick.

Do you work at IOI by any chance and know a guy that goes by the nick-name of sparky ?

I thought that Joker worked on a sports title.
But his dislike towards RSX sure does explain why I haven't heard about any Freedom Fighters, Hitman, or Lynch & Kane for PS3. Though I can't claim to have seen any amazing vertex related in the latter two titles (maybe I'm blind or do not know what VS is or he works on something else).

So we have a reason now! It's the issuing of vertex attributes where RSX is at a disadvantage (actually this info appeared in the original Joker454 thread, but it didn't stick ). This post from SMM explains the problem nicely.

Is this because of Xenos V5 vectors versus RSX's V4? That 5th vector dimension sees a lot of use, and RSX needs to double up fetches/issues to the same info?

Seems like I should dedicate my time to stalking this SMM guy if all his posts are like this and the one in this thread (might even learn something... about police restrainst).

 

[Shifty Geezer]

But with RSX being equivelent to a 7900 GPU for PCs where there's on CPU to help out with VS it would surprise me if the vertex shaders have been trimmed down unless the transistor budget went somewhere else (with VGA engine removed, DirectX instruction set removed).

Some of those things aren't removed because they don't exist. There's no DX instruction set on the GPU for example. DirectX is handled through the driver, which interprets the DX instructions into GPU instructions. It seems to me the VS are as complete as the G70 series, which themselves have comparative limits in what they can do per cycle with vertices versus Xenos (and ATi's other GPUs?)

Now that were are in this stage of debate, to my knowledge unified as in flexible costs more transistors than specialized pipelines compared to the throughput, isn't that true? The way I understand it is that the unified approach is more efficient on open platforms like PCs where the code can't be balanced after the amount of VS and PS pipelines in each GPU, but in a closed system the performance increase isn't as noticable as the code can be tailored for the specialized pipelines (unified will still have an advantage in ease to develop for but not performance).

That's been the argument to date. US is definitely more flexible, but in a closed-box, you can optimize the Bejezus out of the engine. You'll never have the luxury of dedicating 100% of GPU resources to vertex-only work like Xenos, but if that consistitutes only 10% of your overall workload (pie in the sky rhetorical figure), you're not at a huge disadvantage. It depends a lot on the engine. If the vertex-only work is heavily utilized on Xenos, porting that is going to be hard on RSX. Especially when the attributes that can be handled are reduced. A direct port is going to suffer. A well tailored engine looks like it won't.