Could NV30 really be fast enough to do this?
- Thread starter alexsok
- Start date
I'll go out on a limb here, but realtime radiosity is still probably a good ways away (watch as somone proves me wrong!). Maybe really really simple stuff could be done soon, but I've done some stochastic raytracing in the past, and well... I modelled the cathedral of Notre Dame in Chartres. I was using 2 ambient bounces and had fairly high ambient resolution (basically how dense the number of rays sent out is), as the cathedral is a big area, but with lots of detail. On something like 7-8 dual proc machines averaging around 400Mhz for each proc (so maybe 15 400MHz procs), it took about 2-3 days to render a single 1024x768 frame with 6x AA. Now, most scenes probably won't be that big, and I'm sure there are many tricks that can be done to get it going faster on 3D gfx hardware, but I'm doubting we're at the realtime mark quite yet.
Nite_Hawk
Nite_Hawk
Mintmaster
Veteran
Verge said:
Its all about the hacks. You can get pretty close to those kinds of effects even with today's GPU's. I have an idea of how to get similar results, but I need to make a software renderer with GI to generate all the precomputed data I need. Maybe I could find one somewhere...
It would be semi-dynamic, with an object being lit very close to reality as it moves around the room, but not being able to affect the room quite as it should. Boy I wish I had time to implement every graphics algorithm I thought of...
There's no way "true" real-time radiosity is possible yet (and I don't think it will be on silicon-based microprocessors).
I have seen some hacks that use cubic environment mapping for a similar effect, but there's no way they can approach the impact that true real-time radiosity would bring (Similar to how no other shadowing method will match the more-or-less true shadowing that DOOM3 will bring).
As for real-time raytracing, I'd certainly like to see it. There's simply no way to really do good water rendering without raytracing (Components needed for good water rendering: reflection, refraction, angle-dependent reflection/refraction ratio, light attenuation, and lastly lighting of nearby surfaces based on reflection/refraction at the surface of the water). You can do every single one well except the last two by using tricks on today's hardware, and even light attenuation can be approximated fairly well.
As for whether or not it'll be viable, I don't know. I think it would depend on the scene, but I don't see any reason why it couldn't be done, though you wouldn't want to attempt real-time raytracing on scenes that involve lots of reflections and refractions.
I have seen some hacks that use cubic environment mapping for a similar effect, but there's no way they can approach the impact that true real-time radiosity would bring (Similar to how no other shadowing method will match the more-or-less true shadowing that DOOM3 will bring).
As for real-time raytracing, I'd certainly like to see it. There's simply no way to really do good water rendering without raytracing (Components needed for good water rendering: reflection, refraction, angle-dependent reflection/refraction ratio, light attenuation, and lastly lighting of nearby surfaces based on reflection/refraction at the surface of the water). You can do every single one well except the last two by using tricks on today's hardware, and even light attenuation can be approximated fairly well.
As for whether or not it'll be viable, I don't know. I think it would depend on the scene, but I don't see any reason why it couldn't be done, though you wouldn't want to attempt real-time raytracing on scenes that involve lots of reflections and refractions.
Hasn't there been some progress in GI recently using signal processing techniques like wavelets to find radiosity solutions in a domain that is more efficient to calculate?
Anyway, GI will, for the forseeable future, be hacked in via artistry just like it is in CGI rendering right now. I don't know of anyone even using radiosity in prerendered scenes, but instead, writing custom shaders and tweaking diffuse/ambient terms until things "look right"
Anyway, GI will, for the forseeable future, be hacked in via artistry just like it is in CGI rendering right now. I don't know of anyone even using radiosity in prerendered scenes, but instead, writing custom shaders and tweaking diffuse/ambient terms until things "look right"
