Killzone 2 technology discussion thread (renamed)

Status
Not open for further replies.
You would win by being able to push more polygons and shaders. Being able to use many lights is NOT the only benefit of deferred rendering.

If you're in a memory-limited scenario, you would actually lose due to the G-Buffer's inherent existance i.e. less memory to fit the scene geometry.

Also, with deferred lighting, it's better to have many small lights versus fewer large lights (or even down to one world light).
 
AlStrong,

You could very well be correct. But my understanding is that textures consume the vast majority of memory in modern games. Do you know an example of how much memory is consumed by geometry in a modern game?
 
Babcat

The problem is that in a deferred rendering situation, you wind up having to maintain large buffers that retain a lot of data about each pixel that has been determined to be rendered at that point in the display view. Not just Z depth, but material information, texture (color) data and surface normals as well.

It's this significantly larger buffer usage that can cause deferred rendering to be hard in limited memory situations.
 
Babcat

The problem is that in a deferred rendering situation, you wind up having to maintain large buffers that retain a lot of data about each pixel that has been determined to be rendered at that point in the display view. Not just Z depth, but material information, texture (color) data and surface normals as well.

It's this significantly larger buffer usage that can cause deferred rendering to be hard in limited memory situations.

Could you store the buffer in the XDR ram? or is the latency going through the Cell to high, or is there some other bottleneck im not seeing?.
 
Could you store the buffer in the XDR ram? or is the latency going through the Cell to high, or is there some other bottleneck im not seeing?.

You should be able to do so. Writing to main memory is the basis of nVidia's Turbo Cache on PC, and RSX is able to write to XDR at 15GB/s through the FlexIO.

Actually, I wonder if one could write the different render targets simultaneously to each memory pool e.g. write the surface normals into XDR and write the colour data into GDDR3, or if the RTs have to all be in the same memory pool.
 
Actually, I wonder if one could write the different render targets simultaneously to each memory pool e.g. write the surface normals into XDR and write the colour data into GDDR3, or if the RTs have to all be in the same memory pool.

You can not. And it's pointless anyway. Just use XDR for textures.
 
Because you're overall memory limited and not RT memory limited.
So why not just leave RT's in VRAM and offload something else to XDR, like textures.

Well, if it were possible to split the rendertargets, you'd be taking advantage of the two memory buses simultaneously. Deferred shading/lighting is much more memory bandwidth limited because of the render targets, nevermind the space because it's something that's inherent and has to be budgeted before the textures.

If you plan on using the XDR for textures, what difference does it make to the total memory print if you (were able to) split the render target locations? You could just use more memory in the GDDR for textures.
 
Deferred shading/lighting is much more memory bandwidth limited because of the render targets, nevermind the space because it's something that's inherent and has to be budgeted before the textures.

Err...if you have a bus speed problem with RT write, just use more complex shaders.
The whole point of deferred is to balance the shader time/lookup time/write time better.
 
Err...if you have a bus speed problem with RT write, just use more complex shaders.
The whole point of deferred is to balance the shader time/lookup time/write time better.

Of course there's going to be a balancing act. The point I'm trying to convey is making use of the available memory bandwidth. Is that not a form of balancing the work load?

You said yourself you could just store textures in XDR. That's not only making use of storage, but also memory bandwidth (RSX reads XDR at 20GB/s).
 
Any more news about Killzone 2 at Deferred Render/Shading(they are still using 4 SPUs for graphs etc)?

(i dont see any more information since september/07)
 
I have a question, is Guerilla's choice of going down the deferred rendering path, the primary reason why Killzone 2 looks as good as it does?
Ie it allows for complex lighting and high levels of geometry etc.

And with regards to 360 games using full deferred rendering, apparently 360 titles have to use the edram (contains the render output units) so they can't create a G buffer in the 512 main ram, and render from there. So basically while deferred rendering is possible on the 360, there's no advantages in doing so.

So does this put the Xenos architecture at a disadvantage? Or can the 360 match it with optimised forward rendering?
 
if deferred rendering would be the answer to all problems, alot more gamedevs would use it (its not that hard to do). It has both advantages (lights are relatively cheap) and disadvantages obviously, and KZ2 look depends on it to a certain degree. Alot of details are totally unrelated tough, as volumetric effects for example are done on the SPU ( smoke and particles are plenty and look excellent in KZ2 ), so dont base everything on the deffered renderer.

I think its rather the 4 years development time and big budget that heavily factor in how excellent it looks now. Watch 1 year old footage and while looking good, IMHO it dint stand out compared to other titles while it sure as heck it does now.
 
Deferred rendering has a few major advantages and a few minor advantages and since it is a more complicated and expensive process currently, most devs will not go down that path. Since Killzone 2 is first party and been in development a long time they can put in all those resources to make the best of it. Cell processor is also more ideal here for deferred rendering than the processor in 360.
 
Last edited by a moderator:
if deferred rendering would be the answer to all problems, alot more gamedevs would use it (its not that hard to do). It has both advantages (lights are relatively cheap) and disadvantages obviously, and KZ2 look depends on it to a certain degree. Alot of details are totally unrelated tough, as volumetric effects for example are done on the SPU ( smoke and particles are plenty and look excellent in KZ2 ), so dont base everything on the deffered renderer.

I think its rather the 4 years development time and big budget that heavily factor in how excellent it looks now. Watch 1 year old footage and while looking good, IMHO it dint stand out compared to other titles while it sure as heck it does now.

I was also thinking that it's looks are due to more a combination of long development time, a large budget and some great art.

But doesn't DR help with particles? I recall reading deferred rendering allows you to pump out a lot of alpha particles or something.
 
So does this put the Xenos architecture at a disadvantage? Or can the 360 match it with optimised forward rendering?

I think it puts the PS3 at a disadvantage if anything, because most titles are developed on 360 where deferred rendering only brings headaches, I think - I'm not even sure if it is actually budget neutral as you seem to imply.

Also, I still think people seem to overestimate Killzone 2's development time? They probably did some reconnaisance work on PS3 before finishing Liberation on PSP, but it can't be much (or maybe it was shared if anything) and when Killzone 2 production hit off seriously they had a lot of trouble getting their staff - they are based in the Netherlands and there just aren't that many talented and experienced people over here, and I read a newspaper article detailing how they eventually hired and flew people in from all over the globe. Reading up on some interviews, apparently they reached 130 people from 20 nationalities late 2006. That's of course a pretty big team and they've been at it full force for 2 years. So it's still definitely a big budget title!
 
I was also thinking that it's looks are due to more a combination of long development time, a large budget and some great art.

But doesn't DR help with particles? I recall reading deferred rendering allows you to pump out a lot of alpha particles or something.

I think you rarely get as nice a graphical explanation of the different render passes that they combine apparently using an SPE as in this video, that has actually live footage of each of the (7?) different render outputs that are being combined into the final image:

http://www.gametrailers.com/player/43389.html
 
Status
Not open for further replies.
Back
Top