Loading...
> >

Percent Convergence

lorddayradonlorddayradon Posts: 452
edited November 2022 in Daz Studio Discussion

I know I'm using an older driver to get around the nvidia dForce simulation bug. but it seems the % completed of convergence is wierd. I've had to set it to 4, and 99% as some heavy firefly/pixelation in background.

 

8 mins = 75%

11 mins = 85% (700 some iterations)

...and then it stalls.  I end up with 15-20K iterations after a 2-4 hours to get rid of them and not reach target.  I know it is increasingly (exponetionally) hard to get closer to the target the higher the number is. It just seems very disproportionate.

17 mins = 91.6%

and the last 7.4% will take over an hour (or more).

 

Post edited by lorddayradon on

Comments

  • PerttiAPerttiA Posts: 10,024

    What kind of system do you have?

  • lorddayradonlorddayradon Posts: 452
    edited November 2022

    PerttiA said:

    What kind of system do you have?

     

    ryzen 7 5800x 32g. 1070ti oc 

    my 3070ti oc comes tomorrow. And will run both for the cuda  cores.  

    Post edited by lorddayradon on
  • PerttiAPerttiA Posts: 10,024

    Have you checked the log, whether you are rendering on GPU or CPU?

    An 8GB non-RTX card doesn't have much VRAM left for geometry and textures, the 3070ti doesn't improve the situation much (You get 1GB more)

  • lorddayradonlorddayradon Posts: 452

    PerttiA said:

    Have you checked the log, whether you are rendering on GPU or CPU?

    An 8GB non-RTX card doesn't have much VRAM left for geometry and textures, the 3070ti doesn't improve the situation much (You get 1GB more)

    Kinda ambiguious for me to know as I always render using both the vid and cpu. 

    2022-11-14 04:23:35.884 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (NVIDIA GeForce GTX 1070 Ti): 864 iterations, 7.810s init, 127.707s render
    2022-11-14 04:23:35.884 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CPU: 147 iterations, 7.284s init, 127.096s render

     

    Dug back a bit for a big scene, and that appears to be it.

    2022-11-11 05:27:53.364 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (NVIDIA GeForce GTX 1070 Ti): 0 iterations, 6788.013s init, 1867.102s render
    2022-11-11 05:27:53.382 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CPU:                                        31825 iterations, 106.672s init, 227250.516s render

    Well at least I know.

    This raises the question that if I am running both the 3070ti and the 1070ti, and they have a combined memory of  16G, will iRay/Daz not make use of both sets of memory?

  • Richard HaseltineRichard Haseltine Posts: 100,791

    Memory isn't combined (except if you have two identical cards linked with nvLink, and that only for materials).

    Convergence is not linear - it's a measure of how many pixels have stopped changing much (the size of much being set by Render Quality) - well lit with simple materials areas will converge rapidly, areas that are only indirectly lit or have complex materials (e.g. eyes) may be much slower so it is quite possible for a large chunk of the image to be converged while small areas linger for ages.

  • lorddayradonlorddayradon Posts: 452

    So then if I set Daz to use both cards which card's memory will be used or does it use both cards memory independently. 
     

     

  • PerttiAPerttiA Posts: 10,024

    lorddayradon said:

    So then if I set Daz to use both cards which card's memory will be used or does it use both cards memory independently. 

    DS will use both cards if the scene will fit in the VRAM of both of them (independently). When the 1070ti will run out of VRAM, only the 3070ti will render unless that too will run out of VRAM and you are left with snail pace CPU rendering.

  • lorddayradonlorddayradon Posts: 452

    Richard Haseltine said:

    Memory isn't combined (except if you have two identical cards linked with nvLink, and that only for materials).

    Convergence is not linear - it's a measure of how many pixels have stopped changing much (the size of much being set by Render Quality) - well lit with simple materials areas will converge rapidly, areas that are only indirectly lit or have complex materials (e.g. eyes) may be much slower so it is quite possible for a large chunk of the image to be converged while small areas linger for ages.

    this explains a lot 

     

    As a side note both cards have been running very well. Feels like it's cutting the time in half or better. With the faster viewport iray draw and response it even makes scene setup and creation faster.  An image that took approx 120mins at 97% now takes 15mins at 99%

     

    thanx for the help and explanations. 

Sign In or Register to comment.