Digital Foundry Article Technical Discussion Archive [2009]

[psorcerer]

No-one was talking about scalability. That wasn't why Renderman was raised. It's also OT. I outlined in my last line the relevance to this thread of micropolygons, Renderman, etc.

The last line is kind of too generalizing. My point was that there is a limit on how much you can sacrifice everything else to push more polygons, and I suppose it's clear that in case of NG2 a lot of things were sacrificed for "drawing a lot of polygons".

 

[Shifty Geezer]

Those wanting to discuss the new processor can do so here. I tried to move your posts but have no jurisdiction outside the Console zone, so have deleted them instead. But you didn't really say much of interest so what the hey!

 

[Statushock]

IMO, removal of the two armadillos boss battle is more to do with tech limitation of PS3. That particular part is extremely alpha heavy especially when it goes berserk with all the fireworks. Having two of them doing that would be absolute nightmare for RSX. (With IS gone in NGS2 it would be even more nightmare for players to deal with them )

I was wondering about that dual armadillo battle. But what do you think about the Chapter 9 Twin Golden Dragons MazingerDUDE? On Mentor and MN difficulties, they would both be accompanied by squads of lesser black dragons and all of them would bombard the player with intermittent fireballs. The smaller black ones would dive directly at the player at any given time.

Would you say the same for that boss battle? It had so many flying fireballs, many of which were massive.

 

[PeterT]

Pixar's Renderman is the King Of All Renderers. It is highly scalable and efficient, [...]

No-one was talking about scalability.

Ahem.
I think psorcerer's point regarding latency was that just because Renderman is efficient for frame times in the minutes doesn't imply that micropolygon-based rendering is also efficient for frame times measured in milliseconds. (And in my opinion that's a very good point)

 

[grandmaster]

Some self-pimpage here, with the GTAIV time-lapse video I've long been threatening. With each shot taking around 50 minutes to capture this was a bit of a major project...

As the original captures were taken at 1fps, capture tool and encoding overhead only took around 10% of CPU time and a "gargantuan" 200K/s in terms of hard disk throughput (thanks for the most part down to the frame rate and the excellent CineForm HD codec), meaning that the shots could be done in the background without interfering with my work that much.

Taking the captures into Premiere, I found that the software didn't speed up the video enough (it won't speed up any faster than 10000%!), so I rewrote the headers on the captures to 60fps and then found that 500% in Premiere was good enough The final shot is slower, and looks rather impressive so over the Christmas lull when Eurogamer doesn't update, I may do an extended edit that gives a better appreciation of the time of day effects.

 

[Laa-Yosh]

I think psorcerer's point regarding latency was that just because Renderman is efficient for frame times in the minutes doesn't imply that micropolygon-based rendering is also efficient for frame times measured in milliseconds. (And in my opinion that's a very good point)

But we were talking about how rendering subpixel sized triangles via the current hardware rasterization methods is definitely not as efficient as micropolygons, especially if we throw in displacement as well.
Which is why Tim Sweeney, aiming for the next large jump in image quality, believes that maybe software rendering is the way to go in the future.

 

[Richard]

Which is why Tim Sweeney, aiming for the next large jump in image quality, believes that maybe software rendering is the way to go in the future.

Devil's advocate and all but when did Tim NOT say software rendering is the way to go in the future?

 

[Acert93]

Devil's advocate and all but when did Tim NOT say software rendering is the way to go in the future?

With how long Tim has been saying this I wonder if he is taking the stuck clock approach

 

[Laa-Yosh]

Yeah, not that I agree with what Sweeney says, but displacement done right does seem to be a hard challenge for the current GPU architectures.

 

[Silent_Buddha]

Yeah, not that I agree with what Sweeney says, but displacement done right does seem to be a hard challenge for the current GPU architectures.

I think as with anything it's something that will improve over time. Any of the great changes in 3D rendering (wireframe -> gourard shaded -> textured for example) required that someone first tried it then multiple generations of hardware to finally realize it at good speeds and good quality.

Considering displacement mapping + tesselation + sub-pixel polygons are still in it's infancy, WRT realtime rendering for games, I think what has been shown thus far even with the rendering errors (look at early polygon texturing games holy rendering errors batman) has been quite impressive. And as in the past, quality will improve as the hardware catches up to what the 3D artists/engine makers want to do.

And if developers pursue this line (I think it's the next big thing especially with official support in Dx11), we're in for some graphical leaps in the future with regards to realistically rendered environments. All, IMO, of course.

Regards,
SB

 

[liolio]

Just read the last face off, there is something that bother me:

Softening those edges is something that the PS3 can at least make a fairly decent fist of, thanks to the PCF filtering that comes free with the RSX.

I wonder what you mean by free?
I think it could be better to link this link instead of the one in the article, at least people could see different effects at work
I searched in the presentation and actually the artifact in one of the screen looks like the light bleeding problem described in the presentation I linked page 60. On the other hand the noise ( a bit too rough) usually present in the 360 is not visible in this shot.
I wonder what technic they are using on the 360, we have experts here (Marco Salvi and Andrew Lauritzens work appear in the Nvidia presentation, Sebbi ported Nvidia algorithm in TrialHD).
EDIT
I may remember seeing the same kind of noise in one blackrock presentation about oure and slit/second I'll dig that later on, I've to go now.

 

[Laa-Yosh]

Wow, that's some pretty nice character modeling and texture work.

I'll read the analysis later today as well, but for now it seems the game's a little bit of a mixed quality, at least when looking at screenshots on a computer monitor. It might be optimized for HDTVs though...

 

[AlNom]

I wonder what you mean by free?

nV has a hardware accelerated PCF ( I can't recall the number of taps).

 

[semitope]

Wow, that's some pretty nice character modeling and texture work.

The face looks nice, but I think the textures overall in this game are weak. There are some really pitiful 3d models as well in ladders, rails and such.

It would be great if these articles started coming with a bit of background and include some form of information from the developers. It seems this game is still pretty much a 360 game- sub-HD so everything fits in the eDRAM then they go to town. Finding out what they did and plan to do in later games on the PS3 would be good to know.

The jump in textures from mw1 ps3 to mw2 ps3 had me surprised. What really changed and could they take what they did even further?

 

[liolio]

nV has a hardware accelerated PCF ( I can't recall the number of taps).

Interesting, by the look of it and in reagrd to sample in NVDA presentation I would say one tap.
I like the look of it for close objects and when penumbra should not spread on a huge area, on top of it it's cheap

 

[Dave Glue]

The face looks nice, but I think the textures overall in this game are weak. There are some really pitiful 3d models as well in ladders, rails and such.

It would be great if these articles started coming with a bit of background and include some form of information from the developers. It seems this game is still pretty much a 360 game- sub-HD so everything fits in the eDRAM then they go to town. Finding out what they did and plan to do in later games on the PS3 would be good to know.

It's the same 1024x600 res as the other COD's definitely.

 

[joker454]

nV has a hardware accelerated PCF ( I can't recall the number of taps).

Four.

 

[AlNom]

Thanks

 

[Rotmm]

It seems this game is still pretty much a 360 game- sub-HD so everything fits in the eDRAM then they go to town. Finding out what they did and plan to do in later games on the PS3 would be good to know.

I saw you post something similar over at EG. What makes you so sure the game is sub-HD due to the 360 and eDRAM? As the PS3 tends to struggle far more than the 360 with framerate, it seems just as likely to me that the sub-HD decision is so that the PS3 can maintain a fairly decent framerate and that the 360 version is the same resolution to maintain console parity.

 

[semitope]

I saw you post something similar over at EG. What makes you so sure the game is sub-HD due to the 360 and eDRAM? As the PS3 tends to struggle far more than the 360 with framerate, it seems just as likely to me that the sub-HD decision is so that the PS3 can maintain a fairly decent framerate and that the 360 version is the same resolution to maintain console parity.

Its a theory and a question. It was mentioned in the article that the res and 2xaa fit nicely into the edram and I think thats often the case with a choice to go with lower res and usually is coupled with mention of the edram size. For 60fps on the 360 it just works better I think. There are HD games on the PS3 that do 60fps(try), aa and 720p (ratchet and clank ACiT being the most recent and similar example) so my belief is that the difference would be to do with working it on the 360 without tiling. I don't think there is a game like cod and halo on the 360 that run in HD at all if they want to maintain 60fps.

If that argument holds then it suggests to me that the game at least is more tailored for the 360.