Re: [Flightgear-devel] Ray Tracing (was: OpenRT)
On Friday 22 October 2004 00:57, Paul Kahler wrote: > And I thought joining this mailing list would get me > into something different. ;-) > > Ray tracing is not ready for real-time flight simulation. > I'm speaking from experience here. I used to be on the > same mailing list as those OpenRT guys. I also wrote the > renderer behind rtChess: > > http://www.newimage.com/~rhk/rtchess/ > > Which BTW, is available for download (the game) and comes > with docs. Be sure you right-click to drag the board > around. Use the Space-bar for options menu. Unlike most > chess games, you can probably beat this one. > > I've been working the realtime ray tracing (RTRT) stuff > for years, and it's always been "real soon now" for > gaming performance. My library can easily render as many > polygons as you can fit in RAM at a couple FPS (requires > undersampling) on my 700MHz AMD. It's about 40% faster > now than when we wrote rtChess. As with many things, > there is one tiny piece of code that consumes 90% of the > time and I've worked really hard on that. When I get > around to releasing it (probably GPL) some assembly guru > *might* get a bit more performance out of it. > > Because RT is point sampled, the render time is mostly a > function of the number of pixels rather than the number > of polygons. The time complexity is very good WRT scene > size (think log n). A few objects can be handled really > really fast on a modern processor. Unfortunately, using > 1000s of objects brings log(n) up to where things are > rather slow. Going to millions of objects just cuts it > in half again. GPUs are not currently able to handle the > data structures and recursive algorithms required to get > scene scalability, nor do they use double precision. > > Eric Haines once asked me to estimate when RTRT will be > really feasible - I estimated 2012 and so did he. That > doesn't take into account the brick wall we see now with > power density and no 4GHz pentiums and all that... > Multi-core is great for RT. Multi-threading is one of > the features I need to add (done before but removed) > because it scales nearly optimally with more CPUs. > > I could probably (openRT too) load the entire FGFS > scenery database and render it at 1FPS given enough RAM. > Unfortunately that doesn't mean you'll be flying in a > ray traced plane any time soon. > > -Paul interesting stuff. I came into FG from 3d, after moving into that from photography. I've always used Realsoft 3d s/w - currently Realsoft3D V4.5 (Linux Beta) - and while there's a lot of difference in what's being done, there's a lot of cross-over too. One thing that immediately springs to mind is distributed rendering. I guess n/w bandwidth would be a problem but perhaps gigbit eth might cope. LeeE ___ Flightgear-devel mailing list [EMAIL PROTECTED] http://mail.flightgear.org/mailman/listinfo/flightgear-devel 2f585eeea02e2c79d7b1d8c4963bae2d
Re: [Flightgear-devel] Ray Tracing (was: OpenRT)
On October 21, 2004 07:57 pm, Paul Kahler wrote: > And I thought joining this mailing list would get me > into something different. ;-) > > Ray tracing is not ready for real-time flight simulation. > I'm speaking from experience here. I used to be on the > same mailing list as those OpenRT guys. I also wrote the > renderer behind rtChess: > > http://www.newimage.com/~rhk/rtchess/ > > Which BTW, is available for download (the game) and comes > with docs. Be sure you right-click to drag the board > around. Use the Space-bar for options menu. Unlike most > chess games, you can probably beat this one. SDL isn't the only thing that is needed. I've got the following error: error while loading shared libraries: libstdc++-libc6.2-2.so.3: cannot open shared object file: No such file or directory I apt-get the libstdc++6 package, but the error is still there. Ampere ___ Flightgear-devel mailing list [EMAIL PROTECTED] http://mail.flightgear.org/mailman/listinfo/flightgear-devel 2f585eeea02e2c79d7b1d8c4963bae2d
[Flightgear-devel] Ray Tracing (was: OpenRT)
And I thought joining this mailing list would get me into something different. ;-) Ray tracing is not ready for real-time flight simulation. I'm speaking from experience here. I used to be on the same mailing list as those OpenRT guys. I also wrote the renderer behind rtChess: http://www.newimage.com/~rhk/rtchess/ Which BTW, is available for download (the game) and comes with docs. Be sure you right-click to drag the board around. Use the Space-bar for options menu. Unlike most chess games, you can probably beat this one. I've been working the realtime ray tracing (RTRT) stuff for years, and it's always been "real soon now" for gaming performance. My library can easily render as many polygons as you can fit in RAM at a couple FPS (requires undersampling) on my 700MHz AMD. It's about 40% faster now than when we wrote rtChess. As with many things, there is one tiny piece of code that consumes 90% of the time and I've worked really hard on that. When I get around to releasing it (probably GPL) some assembly guru *might* get a bit more performance out of it. Because RT is point sampled, the render time is mostly a function of the number of pixels rather than the number of polygons. The time complexity is very good WRT scene size (think log n). A few objects can be handled really really fast on a modern processor. Unfortunately, using 1000s of objects brings log(n) up to where things are rather slow. Going to millions of objects just cuts it in half again. GPUs are not currently able to handle the data structures and recursive algorithms required to get scene scalability, nor do they use double precision. Eric Haines once asked me to estimate when RTRT will be really feasible - I estimated 2012 and so did he. That doesn't take into account the brick wall we see now with power density and no 4GHz pentiums and all that... Multi-core is great for RT. Multi-threading is one of the features I need to add (done before but removed) because it scales nearly optimally with more CPUs. I could probably (openRT too) load the entire FGFS scenery database and render it at 1FPS given enough RAM. Unfortunately that doesn't mean you'll be flying in a ray traced plane any time soon. -Paul ___ Flightgear-devel mailing list [EMAIL PROTECTED] http://mail.flightgear.org/mailman/listinfo/flightgear-devel 2f585eeea02e2c79d7b1d8c4963bae2d
