New dev tools ? PlayStation 3 Edge

[Tap In]

Au contraire! Remember, my original beef was not that the PS3 was vertex limited, it was that the RSX was vertex limited, which I still stand by 100% today. In fact, I think the article quoted in this thread basically proves it. If you want a real world example, one of my code systems here at work can render ~1.2 million verticies on the 360's gpu faster than the rsx can render the same system with reduced lod models at ~700k verticies. This is all gpu, no cpu help. Yes, the difference is that lopsided in our game, even after I managed to bring that particular system down to two vertex inputs and optimize the heck out of the vertex shader and geometry layout ;( I don't wanna go through that whole thread again like last time, but needless to say to me, rsx still sucks. Your mileage may vary.

I never had a beef with cell though. It has a learning curve, and you need to pretty much rewrite all that code you've been saving over the years, but I think its worth it long term. The new tools and libraries that Sony has been putting out are awesome! I'll wait and see what real world performance we can get before I jump up and down for joy though. We only get 5 spu's full time, and the 6th is "most of the time" since Sony can use it if they need to (the 7th is theirs full time), and I assume with Home and all this other cool stuff they have coming out that they will indeed make use of the 6th.

We're currently fully using 2 spu's, and will most likely use all six by the time we are done since our goal is to hit 60fps on both platforms no matter what. As for how many we'll have free to handle geometry culling, I don't know just yet. I've been having fun with them so far though. I've moved all skinning off the gpu side to cpu side, since we get our biggest gains here on the PS3 by babying the RSX, ie, giving it as little to do as possible to get around its inadequacies. I actually tried it on the 360 as well and it gave back some fps, so the vmx units on the 360 will be fully worked also.

Doing geometry culling on the spu's is something I will do for sure, but I will also try it to a certain extent on the 360 as well, just to see what happens.

thanks for sharing

it's refreshing to hear non-PR stuff about real games.

 

[Shifty Geezer]

Au contraire! Remember, my original beef was not that the PS3 was vertex limited, it was that the RSX was vertex limited...I don't wanna go through that whole thread again like last time, but needless to say to me, rsx still sucks. Your mileage may vary.

Can you explain the difference between Xenos and RSX that results in this? Does RSX have a 'broken' vertex implementation in your opinion (underpowered setup? No inbuilt culling?)? Or does Xenos just have a massive advantage in the US? Is RSX only limited versus Xenos, or versus equivalent GPUs (7800) - is it a PS3 specific design variation? If so, would that be a cost saving thing? Can't see why they'd take an 'off the shelf' GPU and then cobble the vertex units!

 

[Jesus2006]

Au contraire! Remember, my original beef was not that the PS3 was vertex limited, it was that the RSX was vertex limited, which I still stand by 100% today. In fact, I think the article quoted in this thread basically proves it. If you want a real world example, one of my code systems here at work can render ~1.2 million verticies on the 360's gpu faster than the rsx can render the same system with reduced lod models at ~700k verticies. This is all gpu, no cpu help.

Don't wanna dispute, but your observations sound quite differently from what other deveopers here at B3D have stated (just in this very thread) where they pull off 2-3 mio triangles without using the SPUs for vertex shading (on PS3) at all (see HS).

Also your statement that RSX was vertex limited or simply sux was denied several times by others.

 

[Cheezdoodles]

Dunno if this is old or not found some stuff from the GDC presentation

Video


Slides

 

[patsu]

Thanks for posting ! The % improvements mentioned in the slides (e.g., 30% for skinning) ... what are they relative to ?

 

[ShootMyMonkey]

Don't wanna dispute, but your observations sound quite differently from what other deveopers here at B3D have stated (just in this very thread) where they pull off 2-3 mio triangles without using the SPUs for vertex shading (on PS3) at all (see HS).

Well, the numbers he's getting sound low for both platforms, unless he's talking about the scene size and not number of tris actually issued. For instance, I've seen plenty of profiles where we push 3+ million triangles through the pipe at 30 fps on 360 without any real CPU help, but that's including Z-prepasses, shadow maps, bloom passes. The actual size of the scene isn't really 3 million tris.

Also, I wouldn't be surprised if (considering that he mentioned no cpu help), that a lot of the slowdowns he's seeing on RSX aren't really a vertex processing problem, but a vertex issuing problem. i.e. the GPU isn't being kept busy enough.

Also your statement that RSX was vertex limited or simply sux was denied several times by others.

Denied? I don't know about *denied*. The other thread was mainly arguing against the notion that the vertex input limitations are an absolutely insurmountable problem. Joker's posts made it sound as though having to take measures to fight the problem is either impossible or inexcusable or both.

 

[patsu]

Also, I wouldn't be surprised if (considering that he mentioned no cpu help), that a lot of the slowdowns he's seeing on RSX aren't really a vertex processing problem, but a vertex issuing problem. i.e. the GPU isn't being kept busy enough.


Denied? I don't know about *denied*. The other thread was mainly arguing against the notion that the vertex input limitations are an absolutely insurmountable problem. Joker's posts made it sound as though having to take measures to fight the problem is either impossible or inexcusable or both.

Good... I'm not too far off then (I'm learning from [some of] the posts here ! ).

Is culling the only thing Cell helps out RSX in a typical scenario ? How much resources does it take (on the average) ?

 

[Jesus2006]

Denied? I don't know about *denied*. The other thread was mainly arguing against the notion that the vertex input limitations are an absolutely insurmountable problem. Joker's posts made it sound as though having to take measures to fight the problem is either impossible or inexcusable or both.

I was referring (for example) to this statement by nAo.

http://forum.beyond3d.com/showpost.php?p=948227&postcount=42

And this is what joker said:

I don't wanna go through that whole thread again like last time, but needless to say to me, rsx still sucks. Your mileage may vary.

And I wonder what shall be in there that sucks so much [in what is basically a GF 7800]. When asking more detailed questions on that then suddenly some NDAs come up...

 

[homy]

The whole notion that Joker454 described about RSX's vertex limited is useless because Sony discourages developers to do vertex processing/gererating in the RSX.
Until Joker454 moves to the cell, its baseless to state that the RSX is powerful.

 

[Tap In]

I was referring (for example) to this statement by nAo.

http://forum.beyond3d.com/showpost.php?p=948227&postcount=42

And this is what joker said:



And I wonder what shall be in there that sucks so much [in what is basically a GF 7800]. When asking more detailed questions on that then suddenly some NDAs come up...

nAo has no 360 dev kit to compare to, joker does (and he's saying that you NEED the SPUs to match what 360 can do on the GPU side, which leaves a little less available for the Cell to dominate other areas). When you have something to compare it to, you have a different perspective.

it does not mean one or the other is wrong, it's just an opinion based on perspective.

just as StM (cross platform I believe?) said that he has seen other results depending on how things are programmed. Reading betwwn the lines of StM, I'd say he falls somewhere between nAo's and joker's opinions on the matter.

 

[Jesus2006]

nAo has no 360 dev kit to compare to, joker does (and he's saying that you NEED the SPUs to match what 360 can do on the GPU side, which leaves a little less available for the Cell to dominate other areas). When you have something to compare it to, you have a different perspective.

Ok. But isn't Joker saying the exact opposite of what nAo claimed? On the one hand, no vertex limitation and loads of polygons and on the other RSX just sux and needs parts of Cell to compete?

Quite a bit of a difference.

 

[Tap In]

Ok. But isn't Joker saying the exact opposite of what nAo claimed? On the one hand, no vertex limitation and loads of polygons and on the other RSX just sux and needs parts of Cell to compete?

Quite a bit of a difference.

compared to what though?

when running the same game on 360 compared to running it on PS3, you get one perspective with measurable results that allow you to compare the two.

when running it on PS3 only, you get another perspective saying, hey this is doing what we hoped/expected.

it's not an unusual phenomenon

not to say though that given the desire to push the envelope that single paltform devs will most likely be more driven to find work arounds to get the results they want where as in joker's case (multi) it's probably enough of a challenge to figure how to get it to work as well as it was coming from one to the other.

 

[inefficient]

when running the same game on 360 compared to running it on PS3, you get one perspective with measurable results that allow you to compare the two.

The only measurable result running the same code on both is going to show is that running the same code on both is a bad idea.

 

[Tap In]

The only measurable result running the same code on both is going to show is that running the same code on both is a bad idea.

come on, you know what I mean....

see jokers results in the posts above

 

[joker454]

Don't wanna dispute, but your observations sound quite differently from what other deveopers here at B3D have stated (just in this very thread) where they pull off 2-3 mio triangles without using the SPUs for vertex shading (on PS3) at all (see HS).

The numbers I listed weren't for the whole scene, it was just for one of my subsystems. We render far more than 1.2mil verts per frame on each platform! My example used that particular subsystem since I'm intimately familiar with it.

 

[joker454]

Can you explain the difference between Xenos and RSX that results in this? Does RSX have a 'broken' vertex implementation in your opinion (underpowered setup? No inbuilt culling?)? Or does Xenos just have a massive advantage in the US? Is RSX only limited versus Xenos, or versus equivalent GPUs (7800) - is it a PS3 specific design variation? If so, would that be a cost saving thing? Can't see why they'd take an 'off the shelf' GPU and then cobble the vertex units!

Don't really know for sure, thats out of my realm of knowledge. But I do work with someone who does know. Our PS3 lead is insane, he knows Nvidia hardware backwards and forwards. He has tons of Playstation experience, heck he's already shipped two PS3 games! It's one of the reasons I jumped at the chance to be mostly a PS3 guy this time around since I'm learning tons from him. I'll see if I can pull him to this thread on Monday, he can answer alot of this stuff far better than I can. Or then again, maybe it should all be in a new thread? It just occurred to me that this is supposed to be a thread on Edge so I don't want to hijack it anymore than it already has been ;(

 

[Shifty Geezer]

I'll see if I can pull him to this thread on Monday, he can answer alot of this stuff far better than I can. Or then again, maybe it should all be in a new thread? It just occurred to me that this is supposed to be a thread on Edge so I don't want to hijack it anymore than it already has been ;(

Another perspective would be very welcome. One of the issues with people understanding the comments about RSX is no real frame of reference. All we understand is it's a 7800, and to be told it's vertex capabilities are pants seems at odds with the info we have. More info would help explain the apparent large disparity between opinions.

As for if it should be here, I say post here and let the mods decide if it needs a different thread. The vertex capabilities of RSX do tie in with Edge, in a 'is Edge and SPU's doing vertex work essential or a nicety' kind of way. If we understand the limitations of RSX better, we can appreciate the value of Edge moreso.

 

[Tap In]

...One of the issues with people understanding the comments about RSX is no real frame of reference. All we understand is it's a 7800, and to be told it's vertex capabilities are pants seems at odds with the info we have. ...

good post Shifty.

I think the missing ingredient however to understanding "the RSX vertex capabilities are pants" is compared to Xenos as joker had mentioned, perhaps not just compared to 7800.

 

[archangelmorph]

Another perspective would be very welcome. One of the issues with people understanding the comments about RSX is no real frame of reference. All we understand is it's a 7800, and to be told it's vertex capabilities are pants seems at odds with the info we have. More info would help explain the apparent large disparity between opinions.

As for if it should be here, I say post here and let the mods decide if it needs a different thread. The vertex capabilities of RSX do tie in with Edge, in a 'is Edge and SPU's doing vertex work essential or a nicety' kind of way. If we understand the limitations of RSX better, we can appreciate the value of Edge moreso.

I'm not particularly familiar with Cell + RSX at the moment since being new to the industry and starting work on PS2 doesn't give me much of an opportunity (aside from what some of the guys from the team downstairs have told me about them which isn't much because they are always so busy) but I do have a one question...

I find it hard to believe that Sony would make doing all vertex work on the SPUs "essential" for good performance since wouldn't this idle the 8 dedicated vertex pipes on the RSX and leave you with a chunk of the hardware sitting around doing nothing...? I mean, they can't be used for anything else can they..?

Also wouldn't a better solution be to max out the RSX vertex units and then flood any additional work onto the SPEs, leaving more room for extra processing (physics, game logic, sound processing etc..) than if you boycotted using the RSX for this altogether..?

 

[archangelmorph]

good post Shifty.

I think the missing ingredient however to understanding "the RSX vertex capabilities are pants" is compared to Xenos as joker had mentioned, perhaps not just compared to 7800.

But then isn't this a no-brainer considering Xenos has unified shaders?

If the RSX is limited to only 8 vertex units and Xenos has 48 stream processors which could be used for either then correct me if i'm wrong but doesn't that mean that, even if the RSX's vertex capabilities we're no different to that of a 7800 GTX then Xenos could still sh** all over it because it has more/(more flexible) execution units?