Rendering Speed

[super gnu]

ok thats good to know. i was assuming things worked perfectly.. i assume this is also the case for the progressive sampler?

[vlado]

ok thats good to know. i was assuming things worked perfectly.. i assume this is also the case for the progressive sampler?

Yes.

Best regards,
Vlado

[Donfarese]

Super gnu, I rendered both set at 1min 50sec, there is no difference at all between either. Like Vlado said it will keep rendering forever, with a good cap you can render it out at consistent and good times.

[super gnu]

im wondering what the best workflow is to decide what max rays/pixel value to set.. one that is approximately correct(rather, not too much higher than necessary) for a given noise threshold?

[vlado]

im wondering what the best workflow is to decide what max rays/pixel value to set.. one that is approximately correct(rather, not too much higher than necessary) for a given noise threshold?

If you are using the progressive sampler, just leave the max subdivs to 100.

Best regards,
Vlado

[super gnu]

its what i usually do.. or turn it lower.. but with RT? Donfarese was using noise thresh. and max rays/pixel together, and im unsure how to put a reasonable value for rays/pixel for a given noise thresh. normally i just use noise thresh. is it just trial and error?

[vlado]

For RT, that would be 100000 - should work for almost all cases.

Best regards,
Vlado

[3LP]

Max PPP maxes a huge difference in render time for no or nearly imperceptible noise difference.

For my test renders, that you guys can look on my youtube page : https://www.youtube.com/user/sbrusse

Most of my renders are done with noise .005 and 20k as max ppp.

Best way to know the max PPP for me :
1) Set RT GPU as active shade (not production) :
2) open some gpu process manager like evga precision to see when the GPU stops working
3) isolate the most noise bit in you image and crop render ( it can be as small as 20x20 pixels)
4) show statistics ( zoom in as if your crop is too small, you will not see the statistics unless you zoom in)
5) set max PPP to "unlinited" ( like 9999999999 at some stage vray will cap automatically)
6) let the render go, (this can take a while) at some point you will see no difference but EVGA will still show you your GPU is still working and max PPP will go through the roof.
7) printscreen/history save/remember what your noise looks like
set max PPP to 10k
9) crop render somewhere else for 2 seconds and then crop render again where you where right before, or crop render include like 1px more to "refresh" the render so it starts again.
10) look at EVGA precision, once the GPU stops working and the stats don't move, this means you hit either the noise or the max PPP.
11) increase the max PPP by 5k
12) look when the GPU stops and look at the noise
13) if noise is not good enough : go back to 11, if it's good enough/if you don't see any difference/ if it start to render for too long : go to 14.
14) you found a good max PPP for you scene

Depending on the needs, consider if you can't lower a little bit that max PPP as it can cut you render time quite a lot for really only a minimum gain of noise cleanup.

For the rays bundle size 192/8 is the default in GPU but my test have lead me to use 256/64 in stead.
Again a crop render in active shade with statistics will tell what are the best settings to use with your noise/max PPP that you just found out.

If you go too low, your GPU will not be rendering at their max capabilities, CF EVGA precision.

If you go too high, you will force Vray with the "rays per pixel" to sample x rays every-time it make a another pass. IE : Lets go crazy for the theory : if you set 256/256 you will force Vray to sample 256 pixels for each pixels. If noise of the pixel (and or the max PPP) would need only 345 PPP, at the first pass all good it makes 256 PPP for that pixel, but at the second pass, Vray will force 256 PPP for that pixel, making it sample that pixel at 512 PPP even if it only needs 345 PPP. You will over sample that pixel of 167 PPP that doesn't need it = you just waste some render power.

Again, the best way to figure all this out, is to crop render 3-4 (really) small parts in your image to trial/error the best settings by comparing render times in the stats.

This is obviously to squeeze the max power of your GPU, and usually doesn't take more than 5 min to do once you have the concept in mind.

Hope this all make sense, I'm writhing this on my phone so I can make a more elaborated tutorial if needed.

Cheers

Stan