An important visual thing Polyphony have missed
- Thread starter supertone
- 78 comments
- 5,786 views
- 355
- land of the free
- thecoupe
- 205
- New Zealand
- Sh33p
- None
skid marks are definitly needed for improved realism and fun 
and my other little anoyence is that the visual timing of changing gears with the in car veiw is off, he changes gears according to the spedo/ engin revs.. and then the hand moves and physicly changes the gears
and my other little anoyence is that the visual timing of changing gears with the in car veiw is off, he changes gears according to the spedo/ engin revs.. and then the hand moves and physicly changes the gears
Eh.
Plenty of games have done this for a good number of years on the PC...
Cube mapping back in CMR2 had it...
It's not expensive at all (relatively), you just render a 'cube' map in low-res/LOD settings from the car POV, and render it onto your car...
However, this might not be done on other cars (ie, hall of mirrors), only on YOUR car as it was in CMR2...
As per raytracing generally, again, used loads. Doom 3 ran raytracing/self-shadowing, as did PGR2 on the XBox360 for shadows.
Using raytracing for reflections would be costly though, you just render reflection maps/environments every other frame for example... (reflection of car in wet road etc, which we have seen in games loads already, but again, no 2nd reflection (no depth to reflections, just first order))
Dave
Plenty of games have done this for a good number of years on the PC...
Cube mapping back in CMR2 had it...
It's not expensive at all (relatively), you just render a 'cube' map in low-res/LOD settings from the car POV, and render it onto your car...
However, this might not be done on other cars (ie, hall of mirrors), only on YOUR car as it was in CMR2...
As per raytracing generally, again, used loads. Doom 3 ran raytracing/self-shadowing, as did PGR2 on the XBox360 for shadows.
Using raytracing for reflections would be costly though, you just render reflection maps/environments every other frame for example... (reflection of car in wet road etc, which we have seen in games loads already, but again, no 2nd reflection (no depth to reflections, just first order))
Dave
Reflections are easy to compute if it *isn't* important what is actually
reflected. This is the reason why cube-mapping works for reflection
in most 3d scenes. However, if it is important what is reflected, then ...
you need a lot of computing resources at the end of the day, or you have
to leave something out.
Well, how good is a PS3 on raytracing? IBMs interactive Ray Tracer (iRT)
shows off what a PS3 is capable to do when it comes to raytracing.
Have a look at
http://www.alphaworks.ibm.com/tech/irt and draw you conclusion thereof.
A quick overview of the iRT is given in
http://gametomorrow.com/blog/index.php/2007/08/01/off-to-siggraph/.
And don't miss the the video (a realtime raytraced Enzo Ferrari);
http://www.gametomorrow.com/minor/barry/iRT_shaders_560p.mov.
You can download the iRT to your PS3-Linux station from
http://www.alphaworks.ibm.com/tech/irt/download and have some nice
realtime renderings of an Enzo Ferrari (of around 330k triangles) computed
with 6 SPEs. I've done it myself, it runs quite flawlessly.
The iRT on Cell (PS3, BladeCenter QS2X) even beats out the best known
realtime raytracer on a G80 graphics acceleration board due to the Cell
Broadband Engine Architecture (CBEA) and its flexibility in programming.
This is demonstrated here;
http://gametomorrow.com/blog/index.php/2007/09/05/cell-vs-g80/.
The idea behind the iRT is to show off how well the CBEA scales with such
kinds of workloads.
Considering raytracing for GT5(P)....
HW: Given the data/information from all the facts listed above (links), and
from the knowledge of how a raytracer's computational time increases with
scene complexity, can you give an estimate on how many SPEs are required
to render at least 16 (~ 200K tri.) cars within the quality range showing off of
the Enzo Ferrari rendered on a PS3 (using 6 SPEs) via iRT?
reflected. This is the reason why cube-mapping works for reflection
in most 3d scenes. However, if it is important what is reflected, then ...
you need a lot of computing resources at the end of the day, or you have
to leave something out.
Well, how good is a PS3 on raytracing? IBMs interactive Ray Tracer (iRT)
shows off what a PS3 is capable to do when it comes to raytracing.
Have a look at
http://www.alphaworks.ibm.com/tech/irt and draw you conclusion thereof.
A quick overview of the iRT is given in
http://gametomorrow.com/blog/index.php/2007/08/01/off-to-siggraph/.
And don't miss the the video (a realtime raytraced Enzo Ferrari);
http://www.gametomorrow.com/minor/barry/iRT_shaders_560p.mov.
You can download the iRT to your PS3-Linux station from
http://www.alphaworks.ibm.com/tech/irt/download and have some nice
realtime renderings of an Enzo Ferrari (of around 330k triangles) computed
with 6 SPEs. I've done it myself, it runs quite flawlessly.
The iRT on Cell (PS3, BladeCenter QS2X) even beats out the best known
realtime raytracer on a G80 graphics acceleration board due to the Cell
Broadband Engine Architecture (CBEA) and its flexibility in programming.
This is demonstrated here;
http://gametomorrow.com/blog/index.php/2007/09/05/cell-vs-g80/.
The idea behind the iRT is to show off how well the CBEA scales with such
kinds of workloads.
Considering raytracing for GT5(P)....
HW: Given the data/information from all the facts listed above (links), and
from the knowledge of how a raytracer's computational time increases with
scene complexity, can you give an estimate on how many SPEs are required
to render at least 16 (~ 200K tri.) cars within the quality range showing off of
the Enzo Ferrari rendered on a PS3 (using 6 SPEs) via iRT?
- 5,768
- Anoka, MN
Reflections are easy to compute if it *isn't* important what is actually
reflected. This is the reason why cube-mapping works for reflection
in most 3d scenes. However, if it is important what is reflected, then ...
you need a lot of computing resources at the end of the day, or you have
to leave something out.
Well, how good is a PS3 on raytracing? IBMs interactive Ray Tracer (iRT)
shows off what a PS3 is capable to do when it comes to raytracing.
Have a look at
http://www.alphaworks.ibm.com/tech/irt and draw you conclusion thereof.
A quick overview of the iRT is given in
http://gametomorrow.com/blog/index.php/2007/08/01/off-to-siggraph/.
And don't miss the the video (a realtime raytraced Enzo Ferrari);
http://www.gametomorrow.com/minor/barry/iRT_shaders_560p.mov.
You can download the iRT to your PS3-Linux station from
http://www.alphaworks.ibm.com/tech/irt/download and have some nice
realtime renderings of an Enzo Ferrari (of around 330k triangles) computed
with 6 SPEs. I've done it myself, it runs quite flawlessly.
The iRT on Cell (PS3, BladeCenter QS2X) even beats out the best known
realtime raytracer on a G80 graphics acceleration board due to the Cell
Broadband Engine Architecture (CBEA) and its flexibility in programming.
This is demonstrated here;
http://gametomorrow.com/blog/index.php/2007/09/05/cell-vs-g80/.
The idea behind the iRT is to show off how well the CBEA scales with such
kinds of workloads.
Considering raytracing for GT5(P)....
HW: Given the data/information from all the facts listed above (links), and
from the knowledge of how a raytracer's computational time increases with
scene complexity, can you give an estimate on how many SPEs are required
to render at least 16 (~ 200K tri.) cars within the quality range showing off of
the Enzo Ferrari rendered on a PS3 (using 6 SPEs) via iRT?
A cube mapping has to exclude stuff of course
Ie, we will have the target car for the camera only have other cars rendered for it's cube map, the LOD of the cube map render will be pretty low-res to start with, then the LOD of the models will be down too depending upon requirements.
Ie, on PC games at 1600x1200 on-screen res, cars are usually only 1000px long tops (side on), and a ref map of 512x512 usually looks damn impressive!
Lets put it this way, the PS2 had damn fine environment mapping in GT4. Considering the PS3's performance, car in focus reflection of other cars should be a walk in the park from what I have seen.
Yes, having ALL cars reflection each other in full detail (and maybe 2nd level reflections (ie, car reflection in reflection of car)) is a bit unlikely, but the things we'll notice can't be hard!
Dave
Mr Whippy:
I guess, it comes down to fill-rate and texture size problems for now.
The current implementation of the cube-map renderer produces rather
low-resolution textures, these textures are also color compressed (a
drive around London/Suzuka reveals everything). Well, rendering a
higher resolution cube-map would also put more burden on the RSXs
rasterizer. And I guess that the RSXs rasterizer is fill-rate bound, which
can be seen by the frame-drops if cars are close together due to the
transparency effect (which is rather nice), producing a larger portion of
triangle which can't be clipped away. Turning off the transparency effect
would result in a smooth and steady frame-rate, I guess.
However, a 512-cube-map is way too large in memory size. At first
instance, that amount of memory can be put to good use on other things
within GT5(P). If I look at the rendered (on-track) cube-maps, then I guess
that these maps are of order <= 64, whereas the cube-map seen in the
garage might be of order 512.
Using the current implementation of GT5Ps cube-map renderer for
car-to-car reflection would produce a texture that is way too low in
resolution to match the quality standard of the GT brand. The reflected
car(s) would look rather ugly, distracting from the rest of all the other
renderings.
However, I guess we will see car-reflections in GT5. After PD has calculated
all of its texture foodprint/budget for the whole game, and have optimize
all of its rendering stuff, there might be some room left for effects which
are recognized 1% of the times during a race. Of course, I want
car-reflections, too.
I guess, it comes down to fill-rate and texture size problems for now.
The current implementation of the cube-map renderer produces rather
low-resolution textures, these textures are also color compressed (a
drive around London/Suzuka reveals everything). Well, rendering a
higher resolution cube-map would also put more burden on the RSXs
rasterizer. And I guess that the RSXs rasterizer is fill-rate bound, which
can be seen by the frame-drops if cars are close together due to the
transparency effect (which is rather nice), producing a larger portion of
triangle which can't be clipped away. Turning off the transparency effect
would result in a smooth and steady frame-rate, I guess.
However, a 512-cube-map is way too large in memory size. At first
instance, that amount of memory can be put to good use on other things
within GT5(P). If I look at the rendered (on-track) cube-maps, then I guess
that these maps are of order <= 64, whereas the cube-map seen in the
garage might be of order 512.
Using the current implementation of GT5Ps cube-map renderer for
car-to-car reflection would produce a texture that is way too low in
resolution to match the quality standard of the GT brand. The reflected
car(s) would look rather ugly, distracting from the rest of all the other
renderings.
However, I guess we will see car-reflections in GT5. After PD has calculated
all of its texture foodprint/budget for the whole game, and have optimize
all of its rendering stuff, there might be some room left for effects which
are recognized 1% of the times during a race.
Of course, I want car-reflections, too.
- 18
that would be pretty cool... of course its not really important... i think that anything in the game that would make it more realistic or let you customize it to your liking or whatever, like diff paint jobs and stuff, are really cool and make it more fun but arent really a necessity...
sk8rboy1091: you can always play an arcade game if you don't want realism. realism/accuracy is necessary in simulators/simulations.
'i think that anything in the game that would make it more realistic', 'are really cool and make it more fun but arent really a necessity...'. read what you are writing.
'i think that anything in the game that would make it more realistic', 'are really cool and make it more fun but arent really a necessity...'. read what you are writing.
Lain
Premium
- 7,091
- Pasadena, TX
- Yaywalter
Reflections are easy to compute if it *isn't* important what is actually
reflected. This is the reason why cube-mapping works for reflection
in most 3d scenes. However, if it is important what is reflected, then ...
you need a lot of computing resources at the end of the day, or you have
to leave something out.
Well, how good is a PS3 on raytracing? IBMs interactive Ray Tracer (iRT)
shows off what a PS3 is capable to do when it comes to raytracing.
Have a look at
http://www.alphaworks.ibm.com/tech/irt and draw you conclusion thereof.
A quick overview of the iRT is given in
http://gametomorrow.com/blog/index.php/2007/08/01/off-to-siggraph/.
And don't miss the the video (a realtime raytraced Enzo Ferrari);
http://www.gametomorrow.com/minor/barry/iRT_shaders_560p.mov.
You can download the iRT to your PS3-Linux station from
http://www.alphaworks.ibm.com/tech/irt/download and have some nice
realtime renderings of an Enzo Ferrari (of around 330k triangles) computed
with 6 SPEs. I've done it myself, it runs quite flawlessly.
The iRT on Cell (PS3, BladeCenter QS2X) even beats out the best known
realtime raytracer on a G80 graphics acceleration board due to the Cell
Broadband Engine Architecture (CBEA) and its flexibility in programming.
This is demonstrated here;
http://gametomorrow.com/blog/index.php/2007/09/05/cell-vs-g80/.
The idea behind the iRT is to show off how well the CBEA scales with such
kinds of workloads.
Considering raytracing for GT5(P)....
HW: Given the data/information from all the facts listed above (links), and
from the knowledge of how a raytracer's computational time increases with
scene complexity, can you give an estimate on how many SPEs are required
to render at least 16 (~ 200K tri.) cars within the quality range showing off of
the Enzo Ferrari rendered on a PS3 (using 6 SPEs) via iRT?
You, sir, appear to know your stuff. The "homework" is a little too challenging for me to give you a specific answer, but I do know that it does mean it's unlikely we'll see raytracing (at least of that quality) in GT5.
OK I have to say this and some of you may see the relevance and some of you may not? Point being when I was growing up and all we had at the time was BBC Micro's, ZX Spectrum's, Dragon 32K (ooooh), Commodore 16's you where happy enough as a pig in **** as it gave you a form of escape and excitement/enjoyment from an otherwise pretty boring existence (obviously depending on your upbringing) and you didn't moan as you where too busy playing a new game like......ooh let's use Paperboy as an example, now let's get something clear the graphics where by todays standards absolutely ****ing ****, BUT, this in no way took the enjoyment out of the games playability for me as I was too happy with something new to get my melon around, so when something like a PS3 comes along, coupled with an awesome game like GT5 it blows the cobwebs out of the places over machines and titles could not reach!! So to me, a bit of reflection from one car onto another is about as useful as a 3rd wheel on my bike.
Not to sound patronising but kids today just don't know how good they have it
Similar threads
