Upgrading GPU, will this increase visible object count?

flamingspew

New member
I have a huge scene (6 million polys) and at a certain point LW will drop some objects to bounding boxes. The new AMD videocard gets restocked this week. Will 24 GB VRAM mean I can have ~3x the objects (current card is 8GB)?
It’s a big investment, so I want to be sure this will enable more objects.
Thanks!
 
Last edited:

flamingspew

New member
It’s my understanding of OpenGL pipeline that the vectors and textures live on VRAM though. I have 64 GB of RAM and layout does not consume nearly that much before objects go to bounding boxes during layout work. I understand that rendering native uses CPU (hence threadripper).
 

Mastoy

Member
You decide at which count LW will switch to bounding box with the "Bounding box threshold" setting in the display options (shortcut d )
 

next_n00b

Member
It’s my understanding of OpenGL pipeline that the vectors and textures live on VRAM though. I have 64 GB of RAM and layout does not consume nearly that much before objects go to bounding boxes during layout work. I understand that rendering native uses CPU (hence threadripper).
The first thing is, you should specify, which LW version you are using, because <=LW2015 works like 5x faster with Quadro. I assume you are talking about Layout because you mentioned the scene file. My Quadro K6000 with 12GB can easily handle 86 millions polygons in Layout (GPU memory full 100%). I am talking about orbiting/navigating the scene and not deformations of the geometry. Those are done on CPU. 6 millions polygons is like nothing for Layout (orbiting/navigating). And first, check Mastoy advice. Set that parameter to 10kk and see how it plays.
 

Markc

ack ack
I just did a test with a World Creator landscape cloned (8.5mil polys total).
Set bounding box to 10mil, full real time zoom, rotate with an 8gb AMD GPU.
 

flamingspew

New member
Ah thanks! I’m on 2020. I’ll check that bounding box threshold value. I bet that’s it. Good tip! It didn’t get sluggishness at 6m poly.

if i had to bet, geometry deformations are probably single threaded and my threadripper is useless for anything but CPU rendering.

lol now if only bullet was multi-threaded.
 

next_n00b

Member
Seems like it is multi-threaded in lw2020.
 

Attachments

  • multi.jpg
    multi.jpg
    541.7 KB · Views: 59

next_n00b

Member
I think multi-threaded bullet was introducted in 2018? I bought this machine not a long ago to speed up my LW. And what is it now working on 24/7? Mining ETH, it can mine 0.1BTC of ETH per year, electricity deducted. Poor gamers, we stole all GPUs from them.
 

flamingspew

New member
Hah, I guess the computation is just that slow... at least for point-to-point constraints. It’s so slow to me I didn’t even bother opening up my threads.

Anyone know if LW2020 supports the CUDA-enhanced version of bullet? I have an AMD card and CUDA only works on Nvidia cards. That would be worth the upgrade for sure. If you can find one...
 
Top Bottom