Announcement

Collapse
No announcement yet.

Linux and Mac are screwed

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Originally posted by V!NCENT View Post
    Because:
    You're not helping your argument as that can be done without raytracing (even the color-bleeding).

    Originally posted by curaga
    No, I meant reflections.
    You really should go with refractions, because it is one of the few areas rasterization falls flat.

    Comment


    • Originally posted by yogi_berra View Post
      You're not helping your argument as that can be done without raytracing (even the color-bleeding).
      Dynamic and very rich looking atmospheres, without the artist having to apply all kinds of hack all around the place.

      You see, yes you can have all the features that ray tracing has, with traditional rendering techniques, but they are not perfect. They are not pretty. They are complex. They are ugly.

      You see at a certain point you're going to face a serious problem. Not only is the ray tracing algorythm going to be mathemathically faster at a certian point, but once you shoot for higher image quality, where are the ray intersection effects going to be with traditional rendering? How about (let's get absurd; ) you have environmental light, comming in through collored glas, shooting color bleeding in a swimming pool.

      You just can't get the ambience, the feeling of a truely correct image if you do traditional rendering. You just can't.

      Now if you would voxelize the data, you can get all kinds of sick chemical reaction, weather effects, etc. If you're going to render that with traditional rendering (triangle-ized or not), you're simply not getting the right feel.

      ---

      Let's turn the tables. What if (5 years from now or so) you actually can get fully correct ray tracing with voxel data (and all the fun features it can deliver), streaming as unlimited detail? Why would anyone still want to apply traditional rendering techniques? What advantage does it have when physical light is the limit in graphics?

      We are going to hit that bar, sooner or later. There are algorithms out there that are way faster.

      If we can reach the real-time bar. If it's only a matter of time. If it's technologicaly sexy and opticaly correct. If it's enabling artists to not give a shit about the graphics and fully focus on the gameplay and environment... Why the fsck not?

      You really should go with refractions, because it is one of the few areas rasterization falls flat.
      Seriously, not even close... Even reflections (indirect light) falls flatout on its face.
      Last edited by V!NCENT; 20 July 2011, 01:23 PM.

      Comment


      • Originally posted by V!NCENT View Post
        Dynamic and very rich looking atmospheres, without the artist having to apply all kinds of hack all around the place.
        Yeah, that is a load of something. Adding additional render passes is not "applying all kinds of hacks." Nice hyperbole, but not at all accurate.

        You see, yes you can have all the features that ray tracing has, with traditional rendering techniques, but they are not perfect. They are not pretty. They are complex. They are ugly.
        You mean an artist might actually have to learn their tools? For shame. Again it is a nice hyperbole, but not at all accurate. At best you are arguing that you personally cannot create those kinds of effects, which may be true.

        You see at a certain point you're going to face a serious problem. Not only is the ray tracing algorythm going to be mathemathically faster at a certian point, but once you shoot for higher image quality, where are the ray intersection effects going to be with traditional rendering?
        Mathematically faster? What does that even mean? Did you mean computationally faster? Remember, as the CPU power increases, rasterization time decreases as well. You haven't shown otherwise.

        How about (let's get absurd; ) you have environmental light, comming in through collored glas, shooting color bleeding in a swimming pool.

        You just can't get the ambience, the feeling of a truely correct image if you do traditional rendering. You just can't.
        Nonsense. During pre-production of "Finding Nemo" Pixar staff was tasked with recreating a video of a blue whale swimming through the ocean, they recreated it perfectly without raytracing in less than a week. So perfectly in fact that when shown side by side you can't tell which is the rendering and which is the actual video.

        You can see it yourself by watching the additional features on the DVD.

        Let's turn the tables. What if (5 years from now or so) you actually can get fully correct ray tracing with voxel data (and all the fun features it can deliver), streaming as unlimited detail? Why would anyone still want to apply traditional rendering techniques? What advantage does it have when physical light is the limit in graphics?
        What if five years from now I suddenly learn to shit gold bricks? I'll be rich but I'll need a hefty laxative.

        Seriously though, this is a 'what if' that people have been talking about for ever, if it ever happens it'll be a very cold day in Hell.

        If it's enabling artists to not give a shit about the graphics and fully focus on the gameplay and environment... Why the fsck not?
        Yeah right, the more realistic projects become the more time artists have to spend on them to avoid putting people off. Rent 'Beowulf' and you'll see what a rush job will do when you are attempting to replicate anything realistically.

        Even reflections (indirect light) falls flatout on its face.
        No they don't, maybe in your personal experience, but they don't for most people.

        Comment


        • Originally posted by yogi_berra View Post
          Yeah, that is a load of something. Adding additional render passes is not "applying all kinds of hacks." Nice hyperbole, but not at all accurate.
          Well what you're doing is applying drawing tricks. If there is a light shining from the back of your view, oopsy... You'll have to apply pre-rendered ray trace texture. But then your scene is not dynamic. And then when you shine a lamp on a shadowed surface that is actually a texture, then you still see the shadow. Oopsy again.

          Mathematically faster? What does that even mean? Did you mean computationally faster? Remember, as the CPU power increases, rasterization time decreases as well. You haven't shown otherwise.
          There is a point in terms of triangle count that matches the amount of ray calculations needed to be done to get shadows. If you imagine a giant forest with even more detail than Crysis (because that's supposed to be the edge of traingle pushing, for which the game engine is being praised to be able to even run), you're closing in on that limit where ray tracing shadows actually becomes faster. It is the very reason that a lot of new games actually use the blocky ray trace shadows.

          Nonsense. During pre-production of "Finding Nemo" Pixar staff was tasked with recreating a video of a blue whale swimming through the ocean, they recreated it perfectly without raytracing in less than a week. So perfectly in fact that when shown side by side you can't tell which is the rendering and which is the actual video.
          Did they show a rendered image of that with the camera being underwater and looking upwards to the surface of the water? Because perfect refraction isn't going to do you much good if it doesn't exactly mimick the fisheye lense effect. For example if you look up from being deep enough underwater in a swimming pool, you'll notice that not the entire surface of the water actually looks transparant.

          You can see it yourself by watching the additional features on the DVD.
          I don't have it.

          What if five years from now I suddenly learn to shit gold bricks? I'll be rich but I'll need a hefty laxative.
          I thought you weren't going to take that literaly, since it's obvious that Moore's law will meet increasingly faster algorithms up to the point that you no longer need 64 threads to calculate a perfect image.

          Yeah right, the more realistic projects become the more time artists have to spend on them to avoid putting people off. Rent 'Beowulf' and you'll see what a rush job will do when you are attempting to replicate anything realistically.
          For environments, artists can use generators for voxels. Replicating an entire dessert will be a very easy task. Creating entire islands with random number generators for games like Ace Combat will give you the entire terrain. All you need to do is scrap some trees and put it a building or two or a nice bridge and you're done. Airplanes, a runway and a hanger will be all that's left to do for the artists. Even smoke from rockets will be physically calculated. No more particles even...

          No they don't, maybe in your personal experience, but they don't for most people.
          Show me a good caustic and we'll talk...

          Comment


          • Originally posted by V!NCENT View Post
            Well what you're doing is applying drawing tricks. If there is a light shining from the back of your view, oopsy... You'll have to apply pre-rendered ray trace texture.
            Nonsense. It is possible to have lights that do not create shadows. Try again.

            There is a point in terms of triangle count that matches the amount of ray calculations needed to be done to get shadows. If you imagine a giant forest with even more detail than Crysis (because that's supposed to be the edge of traingle pushing, for which the game engine is being praised to be able to even run), you're closing in on that limit where ray tracing shadows actually becomes faster. It is the very reason that a lot of new games actually use the blocky ray trace shadows.
            Citation needed because all of the blocky shadows I've seen are from piss poor shadow mapping. (Crysis isn't the limit of triangle pushing)


            Did they show a rendered image of that with the camera being underwater and looking upwards to the surface of the water? Because perfect refraction isn't going to do you much good if it doesn't exactly mimick the fisheye lense effect. For example if you look up from being deep enough underwater in a swimming pool, you'll notice that not the entire surface of the water actually looks transparant.


            I don't have it.
            Rent it.


            [QUOTE]For environments, artists can use generators for voxels. Replicating an entire dessert will be a very easy task. Creating entire islands with random number generators for games like Ace Combat will give you the entire terrain. All you need to do is scrap some trees and put it a building or two or a nice bridge and you're done. Airplanes, a runway and a hanger will be all that's left to do for the artists. Even smoke from rockets will be physically calculated. No more particles even...[QUOTE]

            Yeah, sure, and then game studios can be drop their art staff to one man in a basement. I'll believe it when I see it.

            Show me a good caustic and we'll talk...
            Again, rent "Finding Nemo."

            Comment


            • Shadows cast by solid objects are actually easier with current rasterisation, mainly because of soft-shadow effects. Ray tracing with soft shadows becomes quite expensive due to the dramatically increased number of rays you have to use.
              Lighting through translucent objects, however, is probably easier with ray tracing. That's not to say that it can't be done with a raster approach, just that it's easier with ray tracing.
              So I'm all for further research with ray tracing (games are not the only application of 3D graphics) and seeing what the strengths and weaknesses of various approaches are.

              I'll also point out that giving artists freedom to design and create however they wish can actually create more work for game development. I was reading somewhere about Rage not likely have to have a vibrant modding community for the reason that content creation is such an effort - a full development team is required, you can't just (easily) do it with a few people.

              As for procedurally created worlds, go visit www.infinity-universe.com.
              Last edited by mirv; 21 July 2011, 11:26 AM. Reason: url correction

              Comment


              • This is a litle bit lame (precomputed ambient oclusion and surface displacement), but this is a video from last year made by nVidia. It shows on average 55 million rays/s at 30fps (55/30=1,83million rays/frame) ray tracing at 1600*1200. It's realtime sparce voxel octree ray tracing 2,7GB of voxel data with CUDA.



                PS: I wonder what GPU that is...
                Last edited by V!NCENT; 30 July 2011, 07:36 AM.

                Comment


                • Originally posted by V!NCENT View Post
                  This is a litle bit lame (precomputed ambient oclusion and surface displacement), but this is a video from last year made by nVidia. It shows on average 55 million rays/s at 30fps (55/30=1,83million rays/frame) ray tracing at 1600*1200. It's realtime sparce voxel octree ray tracing 2,7GB of voxel data with CUDA.



                  PS: I wonder what GPU that is...
                  They also must be using some form of heavy filtering to make the image appear smooth, because 55M rays/sec / 24 FPS / (1600*1200) = Approx 1.2 Sample Per Pixel, which would generally result in a fairly large amount of noise.

                  Although I presume with pre Calculated AO it would be alot quicker.... but isn't the real strength of ray tracing that the shadows/refractions/caustics look great? Having that non dynamic kinda kills the point of ray tracing....

                  A ray tracer that supports all of the above features with quick rendering would be SmallLuxGPU, (sometimes mentioned here) and this is an example:

                  SmallLuxGPU is a LuxRays (http://www.luxrender.net/wiki/index.php?title=LuxRays) demo and a LuxRender (http://www.luxrender.net) "proof of concept".The video...


                  On simple scenes it takes 2-3 seconds for a generally clear render, but can take longer for large interior scenes (outdoors is very fast)

                  The Hardware is 1* 5850 and 2* 5870 and a Core i7

                  I guess if you mix it with faked effects it would be alot faster, but other techniques look good enough for that anyhow.

                  Comment


                  • 4 GPU

                    Note there have been benchmarks run with this software (Based on SLG) that runs on 8* GTX 580's or 4 * 6990's with huge increase in speed:



                    A 5870 gets approx 4000 in score, and 8 580's gets approx 70,000, so it scales across all available devices (via OpenCL) including CPU, but that is almost useless xD

                    Comment


                    • flicking to the end of this thread I have to wonder how on earth a convo on the new win 8 gui ended up here

                      Comment

                      Working...
                      X