I am trying to render a scene that is 10 area lights shining into a parti volume that takes up the entire scene. I am using this as a pass that I will be screening over another image, so I have assigned all the geometry in the scene a black flat color shader. I am also using a physical depth of field camera shader and have my sampling settings set at 0 and 2. My scene is roughly 3000 by 2000. The problem is its taking about 30 hours to render this, which doesnt seem right to me. I am running maya 7 on a 2ghz machine with 2 gb of ram. I have heard that ray marching takes a long time to render, but 30 hours for 1 3000 by 2000 frame seems a little rediculous even with the depth shader attached. My question is, am I doing something wrong? or is this a normal render time for this kind of thing.
thanks in advance.
I'd agree that 30 hrs. seems long..... but you are using a lot of lights with DOF and paritcle volume and a large frame size......all those things take a lot of time... maybe try eliminating one of those factors. Try isolating what exactly is holding up the machine. Turn off the DOF and delete the geomtry and see how long a frame takes. Then try DOF, then up the frame size, then use all ten lights....find out exactly what it's getting hung up on. You can then try to fake stuff or turn down settings appropriately. For example you could fake depth of field in after effects with a BCC plugin-in. All you need is a pass of your animation with a distance from camera based shader going from white to black (found in the maya shaders section of Highend). If it's possible i'd cut back on a light or two as well. It's hard to find what the problem is from here.... but i hoped this helped
A ray marcher represents a whole new magnitude in render time. When we say that raytracing is costly compared to scanline rendering, which is fast enough to do realtime. The comparasion is the same except the fast is now compared to raytracing. (while scan line does 60fps and ray tracing frame in minute so compares raymarching to 36 hours roughly the magnitude yes)
So while complexity grows so grows the rendertime,m, for scanline this time si near linear except each light changes the angle of growth. raytracing is exponential in anture each light causing for a higher exponent, fully raytraced shadow casted fog self shadowing raymarching is SLOW ineeed in comparasion because it grows at a pace of exponentially exponentially, and the growth of the both exponents is very rapid. Granted theres many ways to avoid all this calculation but here a good shader writer is not just worth his weight in gold here, hes worth a unlimited amount of money because he determines the difference between no render and a render when it comes to this.
To make things worse physical DOF makes the fisrts speedy scanline obsolete so it slows things down even more.
And yes if it was a full (selfshadowcasting etc etc) multisampling raymarch i wouldn't be surprised if a 1 hour render would turn into a 30 year one!
So no no amount of cash will enable you to just crank up the most complex of calculations, where even the biggest studios spend years developing ways to cheat because they cant afford to do it otherwise.
And no your not doing anything hugely wrong here, just not udnerstanding that 10 times 10 is muc much less then 10^(10^10)= a number that has 1 and 10 billion zeroes (for linguistic clarity im now using the smaller of he definitions or what the americans sue a s billion, so not to confuse europeans to think its the 1000 larger number) after it! However teh worl is rarely not so bleak with fine tuning you can probably disassociate the exponents to a measly 10^100 whoever by marching lights first you'd drop down this to much much lower levels since you have n secondary march direction.
'
TIP be very very carefully with your partivolumes sample distances as they represent the exponents exponent.
PS this is why such things as brickmaps exsist so you dont need to redoo the raymarch for each fov ray!