The next-gen picture thread 56k warning

[Laa-Yosh]

Your artists would kill you if you'd want them to skin a 2 million polygon mesh. Your GPU would also give up quickly, see my above post...

 

[London Geezer]

I think you'd blow your load on the cat v3 . The rest of the stuff isn't very impressive.



As for unreal 3 tech they may very well be talking about 2 million polygon model . The video makes it sound that way the website does not .


Still if we were to have 10 chars on the screen at once that be 20 million polgons a second not to mention the 100 million used for the city uncompressed .

thats a 120 million polgyons per frame. You'd need to draw that 60 times a second

that be 7.2 billion pps would it not ?

J, remember LOD and culling. You would hardly get 120Mpolys per frame.
As i said, Farcry goes from 30-40Mpolys (no LOD) to around 3Mpolys (default LOD). That's 10 times less.

 

[V3]

Still if we were to have 10 chars on the screen at once that be 20 million polgons a second not to mention the 100 million used for the city uncompressed .

thats a 120 million polgyons per frame. You'd need to draw that 60 times a second

that be 7.2 billion pps would it not ?

Well, not quite, that, 100 million is for the whole city. You can partition the level, since you won't get to see the whole 100 million at once.

But even 7.2 billion pps, PC GPUs will get there soon enough. Hopefully PC will have >4GB of memory by then too.

 

[London Geezer]

Still if we were to have 10 chars on the screen at once that be 20 million polgons a second not to mention the 100 million used for the city uncompressed .

thats a 120 million polgyons per frame. You'd need to draw that 60 times a second

that be 7.2 billion pps would it not ?

Well, not quite, that, 100 million is for the whole city. You can partition the level, since you won't get to see the whole 100 million at once.

But even 7.2 billion pps, PC GPUs will get there soon enough. Hopefully PC will have >4GB of memory by then too.

Well, "possibly" you could have 4GB Ram today. Motherboards support that already (either 3 or 4, can't remember, i think mine is 3...), it's only a matter of going out and buy the Ram modules...
Now, if we're talking about Video Ram, then i expect next gen GPUs to have around 1GB, next next no more than 2GB.

 

[jvd]

Yes u can use a load system but worse case will still be 7.2 billion polygons .

I don't believe the 100 million was for the whole city scape . It may only be per section as the wire mess they show consists of alot of polygons .

But i could stil lbe mistaken.


Think we can take away from this that lod , compresion and other geometry saving tech is a good thing no matter what .

 

[qwerty2000]

Still if we were to have 10 chars on the screen at once that be 20 million polgons a second not to mention the 100 million used for the city uncompressed .

thats a 120 million polgyons per frame. You'd need to draw that 60 times a second

that be 7.2 billion pps would it not ?

Well, not quite, that, 100 million is for the whole city. You can partition the level, since you won't get to see the whole 100 million at once.

But even 7.2 billion pps, PC GPUs will get there soon enough. Hopefully PC will have >4GB of memory by then too.

Well, "possibly" you could have 4GB Ram today. Motherboards support that already (either 3 or 4, can't remember, i think mine is 3...), it's only a matter of going out and buy the Ram modules...
Now, if we're talking about Video Ram, then i expect next gen GPUs to have around 1GB, next next no more than 2GB.

You wish the most VRAM that the next-gen consoles will have is 128MB.

 

[jvd]

London boy. I suspect that the r4x0 and the nv4x will max out at 512 mbs with the r5x0 and nv5x series starting at 512 megs and mabye going to 756 mbs .

It will also be nice to see 3dc gain support as this will also help increase the amount of "virtual" ram that there is .

 

[V3]

Your artists would kill you if you'd want them to skin a 2 million polygon mesh. Your GPU would also give up quickly, see my above post...

Yeah, I thought that could be one of the reason

So I guess we'll be stuck with sub 100,000 poly characters like Dawn, Nalu, or Ruby.

I mean that Unreal characters, could have a base of 100,000 poly with the normal map, should be do able since ATI demoed Ruby already. Why do they go so low to 10,000, which kinda destroyed the look of nice detail they have. Its just kinda shame really.

 

[London Geezer]

Still if we were to have 10 chars on the screen at once that be 20 million polgons a second not to mention the 100 million used for the city uncompressed .

thats a 120 million polgyons per frame. You'd need to draw that 60 times a second

that be 7.2 billion pps would it not ?

Well, not quite, that, 100 million is for the whole city. You can partition the level, since you won't get to see the whole 100 million at once.

But even 7.2 billion pps, PC GPUs will get there soon enough. Hopefully PC will have >4GB of memory by then too.

Well, "possibly" you could have 4GB Ram today. Motherboards support that already (either 3 or 4, can't remember, i think mine is 3...), it's only a matter of going out and buy the Ram modules...
Now, if we're talking about Video Ram, then i expect next gen GPUs to have around 1GB, next next no more than 2GB.

You wish the most VRAM that the next-gen consoles will have is 128MB.

Errrm i meant PC graphics cards, that's what we were talking about.

 

[London Geezer]

London boy. I suspect that the r4x0 and the nv4x will max out at 512 mbs with the r5x0 and nv5x series starting at 512 megs and mabye going to 756 mbs .

It will also be nice to see 3dc gain support as this will also help increase the amount of "virtual" ram that there is .

I really don't know, but i was thinking that so far graphic cards memory has doubled in each generation (more or less)... from 2 to 4, then 8, 16, 32, then we got 64, then 128, now 256... next round is 512, then i think we'll just go 1GB. But i am really speculating at this point...

 

[Laa-Yosh]

So I guess we'll be stuck with sub 100,000 poly characters like Dawn, Nalu, or Ruby.

I mean that Unreal characters, could have a base of 100,000 poly with the normal map, should be do able since ATI demoed Ruby already. Why do they go so low to 10,000, which kinda destroyed the look of nice detail they have. Its just kinda shame really.

Nalu and Ruby don't have 100K polys in their original form either - they are subdivided using the Catmull-Clark scheme, I'd say 2 iterations. The artists have modeled an ~10K control mesh, UV mapped it, skinned it to a skeleton, and then had the polygons subdivided and smoothed by the 3D app. For animation, it happens for each frame before rendering, thus the subdivision will also smooth out any smaller skinning errors as well.

It is actually a pretty standard workflow and I see no reason to interfere with it. If you need more detail, use displacement mapping - it is far more efficient for both the production pipeline, and rendering on the GPU as well.

 

[qwerty2000]

The ps3 can't do that golf picture that I put up, you guys are right. But the rest of the pictures I put up the ps3 should be able to do it.

 

[jvd]

The ps3 can't do that golf picture that I put up, you guys are right. But the rest of the pictures I put up the ps3 should be able to do it.

so will only the ps3 be able to render those pictures u put up or will all next gen systems be able to do so ?

 

[V3]

Nalu and Ruby don't have 100K polys in their original form either - they are subdivided using the Catmull-Clark scheme, I'd say 2 iterations. The artists have modeled an ~10K control mesh, UV mapped it, skinned it to a skeleton, and then had the polygons subdivided and smoothed by the 3D app. For animation, it happens for each frame before rendering, thus the subdivision will also smooth out any smaller skinning errors as well.

The Unreal characters, the high res model, they were subdivided as well right ? or other HOS. I mean no one really tweak 2 million polygons models. I mean most likely the model is coverted to polygons just to generate the normal maps.

It is actually a pretty standard workflow and I see no reason to interfere with it. If you need more detail, use displacement mapping - it is far more efficient for both the production pipeline, and rendering on the GPU as well.

I understood that, what I am saying is why do they go to low 10,000 polygon ? I mean if they are going to displace this base model back into 2,000,000 I won't complain, but for normal map, I would rather they used a more detail base.

 

[London Geezer]

Nalu and Ruby don't have 100K polys in their original form either - they are subdivided using the Catmull-Clark scheme, I'd say 2 iterations. The artists have modeled an ~10K control mesh, UV mapped it, skinned it to a skeleton, and then had the polygons subdivided and smoothed by the 3D app. For animation, it happens for each frame before rendering, thus the subdivision will also smooth out any smaller skinning errors as well.

The Unreal characters, the high res model, they were subdivided as well right ? or other HOS. I mean no one really tweak 2 million polygons models. I mean most likely the model is coverted to polygons just to generate the normal maps.

It is actually a pretty standard workflow and I see no reason to interfere with it. If you need more detail, use displacement mapping - it is far more efficient for both the production pipeline, and rendering on the GPU as well.

I understood that, what I am saying is why do they go to low 10,000 polygon ? I mean if they are going to displace this base model back into 2,000,000 I won't complain, but for normal map, I would rather they used a more detail base.

I personally think they lowered it to 10K just to show off how good they can make the models look, even when they're so "low" poly compared to the original. It's not indicative of the amount of polygons there will be on the models used for games based on the Unreal Engine 3. I think it as a way of saying "Hey, look how cool this looks, AND IT'S ONLY 10K POLYGONS!"

 

[qwerty2000]

The ps3 can't do that golf picture that I put up, you guys are right. But the rest of the pictures I put up the ps3 should be able to do it.

so will only the ps3 be able to render those pictures u put up or will all next gen systems be able to do so ?

I think the xbox2 could do it but I think the N5 will have a hard time.

 

[jvd]

The ps3 can't do that golf picture that I put up, you guys are right. But the rest of the pictures I put up the ps3 should be able to do it.

so will only the ps3 be able to render those pictures u put up or will all next gen systems be able to do so ?

I think the xbox2 could do it but I think the N5 will have a hard time.

i doubt any of them can do it

 

[London Geezer]

The ps3 can't do that golf picture that I put up, you guys are right. But the rest of the pictures I put up the ps3 should be able to do it.

so will only the ps3 be able to render those pictures u put up or will all next gen systems be able to do so ?

I think the xbox2 could do it but I think the N5 will have a hard time.

Why exactly, apart from pure speculation on your part, that is.

 

[Laa-Yosh]

The Unreal characters, the high res model, they were subdivided as well right ? or other HOS. I mean no one really tweak 2 million polygons models. I mean most likely the model is coverted to polygons just to generate the normal maps.

Yep, they seem to have 2-3 iterations of subdivision applied. Note that most applications do not support fully implemented subdivision surfaces (tesselated to infinite smoothnes), but rather only apply the subdivision scheme to polygon models, thus generating more polygons. Renderman works differently; Maya subdivs are also different because they have adjustable tesselation, more like NURBS.

By the way, there's a tool to work with insanely detailed meshes, called Zbrush (www.pixologic.com).

I understood that, what I am saying is why do they go to low 10,000 polygon ?

Because the engine uses stencil shadows for dynamic lights (according to Daniel Vogel's post), and higher poly counts would kill performance.

 

[qwerty2000]

The ps3 can't do that golf picture that I put up, you guys are right. But the rest of the pictures I put up the ps3 should be able to do it.

so will only the ps3 be able to render those pictures u put up or will all next gen systems be able to do so ?

I think the xbox2 could do it but I think the N5 will have a hard time.

i doubt any of them can do it

You're underestmaing the next-gen consoles, none of thoes screens goes above 1bbps. With the right rendering technqiues and gemotry we'll be see this kinda of graphics in the next gen.