Chaperone

[Guden Oden]

By the way, it's Paula. AGNUS and Denise, if you please. Angus is a dewd's name and these three are all ladies.

Of course, post-Amiga 1000, Agnus put on some weight and became Fat Agnus... She's still a lady tho.

 

[j^aws]

How well could the Gigapixel architechture pull off real time displacement mapping?

Dunno, but here are some nVidia TBDR related patents,

Demand-based memory system for graphics applications

Optimized system and method for binning of graphics data

circuit and method for processing render commands in a tile-based graphics system

 

[j^aws]

How well could the Gigapixel architechture pull off real time displacement mapping?

Dunno, but here are some nVidia TBDR related patents,

Demand-based memory system for graphics applications

Optimized system and method for binning of graphics data

circuit and method for processing render commands in a tile-based graphics system

Here's another one,

Rendering pipeline

 

[one]

This article is a bit old, but that's about all that's known about the PS3 GPU. They say it will be a sort of combination of NV40 and NV50

Babelfish is misleading for the article...

 

[Titanio]

still don't have any kind of meaningful specs about the GPU

Sure they do...
it's a... uh... faster then a 6800... and stuff...

Tease

I guess this at least confirms it'll be north of 400 Mhz...

My biggest question about the PS3 GPU is: is it dedicated solely to pixel shading, or is vertex shader also on the card? Or is it more heavily weighted toward pixel shading? I'm under the impression that assuming this was the case, with vertex processing on Cell, and assuming the GPU was of similar size in terms of logic etc. and of similar clockspeed etc. to the competition (R500), it should/would hand PS3 a greater and more visually appreciable advantage in terms of graphics capabilities. Does everyone tend to agree with that, or are there holes in the argument?

 

[j^aws]

...
My biggest question about the PS3 GPU is: is it dedicated solely to pixel shading, or is vertex shader also on the card?
...

I'm going with TBDR at this moment. A similar setup to ELAN + PowerVR, i.e. CELL + G7x.

The CELL would act as the geometry engine and the G7x act as the deferred fragment shader.

Unless Dave gives any kind of sign, I'm going to build a 'voodoo doll' and prepare a time lapsed 'curse' to unleash untold wrath and bad karma Dave's way!

 

[Titanio]

...
My biggest question about the PS3 GPU is: is it dedicated solely to pixel shading, or is vertex shader also on the card?
...

I'm going with TBDR at this moment. A similar setup to ELAN + PowerVR, i.e. CELL + G7x.

The CELL would act as the geometry engine and the G7x act as the deferred fragment shader.

Unless Dave gives any kind of sign, I'm going to build a 'voodoo doll' and prepare a time lapsed 'curse' to unleash untold wrath and bad karma Dave's way!

Interesting, I hadn't considered the possibility before.

Do you think it's possible it could be simultaneously a TBDR and a variant of NVidia's next-gen PC line, if the next-gen PC line isn't TBDR? Assuming no extra preprocessing hardware (preprocessing in software on Cell), how much work would be required to get the pixel shaders working on a per-tile basis?

Do you think they'd include specialised hardware for scene preprocessing, or could Cell handle that handsomely? If they did have to include specialised hardware, what kind of silicon budget might be required? In the end, if I understand things correctly, TBDR eliminates unnecessary work for the pixel shaders, but you'd still need pixel shaders at the end, right? Would the inclusion of dedicated preprocessing hardware eat into your pixel shading logic budget much? Presumably if it did, that would be offset by a reduction in redundant shading/texturing? (I'm probably REALLY showing my ignorance here - do current GPUs shade unnecessarily?)

Also, could this potentially allow for even better quality AA over competing products? Reading early articles about PowerVRs TBDRs versus NVidia and 3DFX's offerings, it seems it did hold an AA advantage, but would that still hold true today?

 

[pc999]

Would a "like a 6800 ... but faster" be able to that

 

[Titanio]

Would a "like a 6800 ... but faster" be able to that

Now now, let's not rewrite what Faf said:

"it's a... uh... faster then a 6800... and stuff..."

nowhere did he say it was "like a 6800" and there's always that "and stuff" bit

Hehe, seriously, I don't think TBDR is likely (it isn't..is it?), but I'm interested in potential advantages and disadvantages for PS3 from an academic viewpoint, I guess.

 

[j^aws]

...
My biggest question about the PS3 GPU is: is it dedicated solely to pixel shading, or is vertex shader also on the card?
...

I'm going with TBDR at this moment. A similar setup to ELAN + PowerVR, i.e. CELL + G7x.

The CELL would act as the geometry engine and the G7x act as the deferred fragment shader.

Unless Dave gives any kind of sign, I'm going to build a 'voodoo doll' and prepare a time lapsed 'curse' to unleash untold wrath and bad karma Dave's way!

Interesting, I hadn't considered the possibility before.

Do you think it's possible it could be simultaneously a TBDR and a variant of NVidia's next-gen PC line, if the next-gen PC line isn't TBDR? Assuming no extra preprocessing hardware (preprocessing in software on Cell), how much work would be required to get the pixel shaders working on a per-tile basis?

No, both next gen PC and PS3 would have to be based on the same next gen tech. Dunno on your other question...

There are rumours that NV do not like unified shaders and that they are going multi chip setups with dedicated VS and PS chips in the future, i.e. components of a current GPU on separate chips. So their VS chip would be the equivalent of the aforementioned ELAN chip.

For PS3, the question still remains whether their geometry engines, i.e. VS units/chips work can be replicated on CELL. So nothings really changed here!

Do you think they'd include specialised hardware for scene preprocessing, or could Cell handle that handsomely? If they did have to include specialised hardware, what kind of silicon budget might be required? In the end, if I understand things correctly, TBDR eliminates unnecessary work for the pixel shaders, but you'd still need pixel shaders at the end, right? Would the inclusion of dedicated preprocessing hardware eat into your pixel shading logic budget much?

Dunno about silicon budget but all specialised deferred/fragment/pixel processing logic would be on the G7x. Leave all general purpose geometry related work for CELL...

Also you'd save transistors elesewhere, i.e. TBDR would reduce your need for framebuffer bandwidth and RAM...

Also, could this potentially allow for even better quality AA over competing products? Reading early articles about PowerVRs TBDRs versus NVidia and 3DFX's offerings, it seems it did hold an AA advantage, but would that still hold true today?

Well the architecture is designed to save work for better AA. However, the R500/ Xenon setup also looks like it was built around optimising AA...

 

[pc999]

Would a "like a 6800 ... but faster" be able to that

Now now, let's not rewrite what Faf said:

"it's a... uh... faster then a 6800... and stuff..."

nowhere did he say it was "like a 6800" and there's always that "and stuff" bit

Hehe, seriously, I don't think TBDR is likely (it isn't..is it?), but I'm interested in potential advantages and disadvantages for PS3 from an academic viewpoint, I guess.

I am not talking about Faf but this

http://www.beyond3d.com/forum/viewtopic.php?p=483717&highlight=#483717

The graphics chip isn't there, say, but we can get a pretty good idea by taking an NVIDIA 6800 and saying, okay, it'll be like this but faster."

 

[Titanio]

Would a "like a 6800 ... but faster" be able to that

Now now, let's not rewrite what Faf said:

"it's a... uh... faster then a 6800... and stuff..."

nowhere did he say it was "like a 6800" and there's always that "and stuff" bit

Hehe, seriously, I don't think TBDR is likely (it isn't..is it?), but I'm interested in potential advantages and disadvantages for PS3 from an academic viewpoint, I guess.

I am not talking about Faf but this

http://www.beyond3d.com/forum/viewtopic.php?p=483717&highlight=#483717

The graphics chip isn't there, say, but we can get a pretty good idea by taking an NVIDIA 6800 and saying, okay, it'll be like this but faster."

Ah, I see. It's difficult to reach conclusions based on that comment, without a bit of context. As the article states, no graphics chip was in the kit at that stage, so the weight of that comment would depend on whether Sony had given any information to the developer about the GPU or not. If not, that could simply be their guess or expectation rather than SCEI-informed thought.

However, I too expect it will be somewhat like taking a 6800 (in that it'll still be SM3.0+, though maybe a little more +) and upping the performance. Hopefully they'll be removing the vertex shading silicon and expanding the pixel shading silicon, though..

Also, I'm beginning to wonder if Faf's post was a sarcastic reference to that developer comment in that gamesindustry article Did Faf mention having PS3 kits before? I think I remember him saying some months ago now that they were expecting them shortly..

Oh, and cheers for the answers Jaws. My guess is if it has to be tied so closely to the PC line, it might reduce the chances of a TBDR, but who knows!

 

[j^aws]

...
Oh, and cheers for the answers Jaws. My guess is if it has to be tied so closely to the PC line, it might reduce the chances of a TBDR, but who knows!

Why would that matter if the PS3 GPU/ G7x is the next gen PC part but customised/ tailored for PS3 anyway?

The only problem I see is if it doesn't fit in with their WGF 2.0 roadmap...

Btw, Dave, it is ready...

 

[Titanio]

...
Oh, and cheers for the answers Jaws. My guess is if it has to be tied so closely to the PC line, it might reduce the chances of a TBDR, but who knows!

Why would that matter if the PS3 GPU/ G7x is the next gen PC part but customised/ tailored for PS3 anyway?

Well, yeah, that's why I was wondering how much work it'd take to tailor a non-TBDR architecture as a TBDR for PS3..not sure myself. WGF 2.0 is a factor though..

Also, isn't G7x a mobile series, supposedly? PS3 would be a G80 variant?

And yeah, I'm tempted to start making my own voodoo dolls soon

 

[3dcgi]

do current GPUs shade unnecessarily?

Yes. That is why we have the technique of rendering the entire scene without shading enabled to create a final z buffer. This way pixels that don't need to be shaded are rejected by the early z hardware and no unecessary shading is performed.

 

[ndoogoo]

do current GPUs shade unnecessarily?

Yes. That is why we have the technique of rendering the entire scene without shading enabled to create a final z buffer. This way pixels that don't need to be shaded are rejected by the early z hardware and no unecessary shading is performed.

Its all done for you on PowerVR hardware Shame the PowerVR 5 never made it into any of the next-gen consoles.

Atari's Jaguar console had the best chip names, Tom & Jerry

 

[Ty]

Atari's Jaguar console had the best chip names, Tom & Jerry

Consoles? Ok. All time? Would have to give that to the Amiga with Fat Agnus.

 

[j^aws]

...
Also, isn't G7x a mobile series, supposedly? PS3 would be a G80 variant?
...

The G70 and G80 are *allegedly* new internal codenames for nVidia's next generation GPUs. And they would skip the rumoured cancelled NV50.

As I mentioned earlier in the thread, I'm guessing the '70' and '80' internal designations would fall inline with the '6x00' series external designations.

As for the 'G', the change from 'NV', NVidia tech, to 'G' for Gigapixel tech, makes the most sense to me for a new architecture that's not based on any existing architecture from NV. If only because it adds more fuel to the TBDR fire!

 

[nAo]

As for the 'G', the change from 'NV', NVidia tech, to 'G' for Gigapixel tech, makes the most sense to me for a new architecture that's not based on any existing architecture from NV.

G as GeForce makes even more sense to me

 

[j^aws]

As for the 'G', the change from 'NV', NVidia tech, to 'G' for Gigapixel tech, makes the most sense to me for a new architecture that's not based on any existing architecture from NV.

G as GeForce makes even more sense to me

Haha...I know but GeForce is an *external* marketing name, why bring that *internal*? It's like forgetting your name and labelling your clothes! It would make more sense just to keep 'NV'!