Animation: saved IRR map takes time during rendering?

[Lupaz]

Hi,

Although this was probably discussed many times, I couldn't find the answer to this question:

If I saved the IRR map + LC, why is there such a big difference in render time if I have GI checked compared to un-checked?
I thought that, once I have the GI in a saved map, the GI itself wouldn't take much time during rendering.

Is there any parameter that controls the way DMC uses the saved GI solution?

Thanks!

[Lupaz]

Anyone?
thanks.

[super gnu]

interp. samples has quite a big effect on rendertime, vray still has to do a reasonable amount of work filtering the map before use. reducing that can speed up the render at the cost of introducing blotchies if your imap isnt clean enough.

its hard to say what might be the issue though, you dont mention what sort of difference in rendertime you are getting.

normally rendering without gi will always be a bit faster, even compared to using saved maps. for instance without gi you will normally have many darker areas in your scene, which vray will naturally spend less samples on.

[Lupaz]

I'd say the difference is huge. BUT, thanks! You were totally right.
I'm attaching 3 images.
The first one with pre-saved GI with int samples at 50.
The second image with GI off.
The third image with pre-saved GI with int samples at 1.

The render time is on the stamp on each image.

Thanks again.

[Neilg]

it looks like you're saving all of your lights with irradiance map too - on something as complex as a tree that will increase the time a lot. In your version with gi off your lights arent doing anything - if you want an accurate comparison you need to switch that off.

[super gnu]

interpolation samples of 50 is extremely high.. default is 20, and i usually try to go down from there, not up.

also if its trees you are rendering, firstly you wont see the blotchies if there are any, so no need for a high interp. setting, and secondly there will be a crapload of imap samples to interpolate, so it will have to do a load of work with 50 interpolation samples.


btw are you doing a still/ flythrough or animation mode type job? if youre using animation mode, then interpolation samples has an even more massive effect on rendertime, and also, should be set lower the more imap frames you are interpolating.

[super gnu]

cubiclegangster - is it really the case that using "store with irradiance map" is significantly slower wrt the imap? youd probably get more samples in the imap i guess, but its always been far preferable for me to use a very detailed imap with this option than to try to render loads of raytraced lights.

or were you just referring to the fact his no gi render also had no lights so would obviously go unrealistically fast.

[super gnu]

ive also notived a very drastic difference between the images with interp. samples at 50 and at 1. since ive never tried 1 interpolation sample maybe its screwing something up.. worth trying a reasonable setting of 5-10 and seeing how it goes? you shouldnt get a major change in look by changing the interp samples.

[Lupaz]

It is an animation btw. The camera is going forward in this case.

Hi Cubiclegangster. I use my lights in irr map mostly to avoid noise and speed up rendering.

super gnu: I saved the imap with incremental add and then from file, like in the old days I guess.
I used 1 interp sample just to see how much that would affect the render time. I didn't try the animation with lower values yet, but I'm thinking using about 15 to 20.

I AM very surprised with how much the render times go up. I had no clue.

[super gnu]

id suggest if its just trees you likely can get away with a lower setting that 15-20.. as long as you dont have that same issue as with 1 sample, where the gi looks totally different. - im still assuming that change is being caused by the specific setting of 1 sample.. i.e. no interpolation at all.

as a plus, with lower interp samples the gi will look sharper too. 50 samples will blur the crap out of it.

[Lupaz]

I wonder if there's a way to just bake the entire GI...So the info (color) would be already on the geometry or existing shader. This way you can get the effect of GI with low render times and no noise.

Is there? Or this would be the same as having no interpolation?

[Neilg]

or were you just referring to the fact his no gi render also had no lights so would obviously go unrealistically fast.

It was this one - it's not a fair test, the 'no GI' test that was done wasnt a true comparison of render times because the lights were effectively off.

[super gnu]

i think the only way to do what you are asking is texture baking , which is a whole load of faffing round.. youd render the whole thing in the time it took you to prepare everything and bake the lighting to textures.

to be honest with renderpower so cheap these days people often dont even bother with the imap any more, just brute force it. im still an imap fan personally. of course if youve got significant animated geometry then its brute force and lightcache all the way.


could always use radiosity in max to bake the lighting then youd have some nice fast renders.

- first jobs we ever did with vray were done like this, the imap was still a bit flaky back then, and we were very comfortable in lightscape.. so we did the lighting in lightscape then the materials and rendering in vray. worked very well iirc, and wickedly fast. wouldnt like to use lightscape/radiosity on trees though!


got me thinking.. id love to see how fast lightscape would go on a modern machine. it was hypnotic to watch it building up the lighting face by face.

[Lupaz]

these days people often dont even bother with the imap any more, just brute force it.

I just did a test with BF, out of curiosity.
Not good. Takes too long. I'm gonna stick with the good old Irr map for now.

[Neilg]

you may already be doing this, but are you rendering your base image with 'visible to camera' switched off on the trees, and rendering them in a separate pass?