Hello. I am always indebted.
I am currently conducting research using optix6.0.
I need to use hundreds of millions of triangular polygons.
The calculation is performed using the class. I understand that optix can handle about 10 million to 30 million polygons at a time.
Therefore, we calculate by dividing hundreds of millions of polygons. I thought that this could be done by updating the polygon information in the buffer every time. However, doing so will result in memory errors.
So I discard every time. However, it takes time to destroy .
Is there any way to update the polygon information in the buffer without destroying the ?
Hi CherryBlossoms,
OptiX 6 handles memory allocation and deallocation at launch time. When you need OptiX 6 to deallocate things, you can avoid the need to destroy your context by running a “dummy” launch of size 1x1 with an empty raygen program, after you make changes to your scene, but before you allocate new memory. This way you can avoid needing to have enough memory to store both large meshes at the same time.
It is easier to control memory with OptiX 7, so for a project with large amounts of memory it might be worth considering whether to use OptiX 7 if you can.
–
David.
Hello, David.
Thank you for the proper answer. I am always helped by you.
The question changes a lot, but let me ask you another question.
I think Optix uses a bounding box and its tree structure to speed up processing. Is that done on the GPU side?
Also, does it include recursive processing?
Hi CherryBlossoms,
Yes, OptiX uses a Bounding Volume Hierarchy (BVH) tree to speed up intersection tests. This is done on the GPU, and yes it does include recursive processing. This recursive processing is what we call “traversal” because it’s traversing the tree structure. With RTX cards, recursive traversal is performed in hardware, so that it’s very fast. When using non-RTX GPUs, the recursive processing is done with CUDA on the GPU. The BVH tree structure is also created on the GPU.
–
David.