Houdini Docs(Source: https://www.sidefx.com/docs/houdini/vellum/shapematching.html)
Shape match simulations require the latest driver for NVIDIA Ampere (or later) cards. Older chipset architectures aren’t supported. If you experience problems, go to Houdini’s main menu. There, open Edit ▸ Preferences ▸ Miscellaneous and set OpenCL Device to CPU. Then, restart Houdini to apply the changes.
I'm using RTX 3070, which is of Ampere architecture, but with "OpenCL Neighbor Search" enabled in Vellum Solver, the operator throws plenty of errors. Notably:
Invalid source /obj/geo1/vellumsolver1/dopimport_geometry/load_each_field_end
Error: The sub-network output operator failed to cook: /obj/geo1/vellumsolver1/dopimport_geometry/switch3
Error: Invalid source /obj/geo1/vellumsolver1/dopimport_geometry/load_a_field
Error: vellumobject1 - /obj/geo1/vellumsolver1/dopnet1/vellumsolver1/graph_color_constraints: Error: Vex error: The number of points in the geometry and constraints do not match.
vellumobject1 - /obj/geo1/vellumsolver1/dopnet1/vellumsolver1/ocl_neighborsearch_grains: Error: /obj/geo1/vellumsolver1/dopnet1/vellumsolver1/ocl_neighborsearch_grains/neighborsearchcl1: OpenCL neighbor search supports only uniform radius particles.
vellumobject1 - /obj/geo1/vellumsolver1/dopnet1/sources/defaultsource/single_emission: Error: Invalid source /obj/geo1/vellumsolver1/dopnet1/sources/defaultsource/single_emission/dop_geometry2
Error: vellumobject1 - /obj/geo1/vellumsolver1/dopnet1/vellumsolver1/graph_color_constraints: Error: Vex error: The number of points in the geometry and constraints do not match.
vellumobject1 - /obj/geo1/vellumsolver1/dopnet1/vellumsolver1/ocl_neighborsearch_grains: Error: /obj/geo1/vellumsolver1/dopnet1/vellumsolver1/ocl_neighborsearch_grains/neighborsearchcl1: OpenCL neighbor search supports only uniform radius particles.....
As a result, geometry is disappearing on every other frame and the simulation doesn't solve at all.
To solve it I have to either disable "OpenCL Neighbor Search", which most likely slows down calculations, or use uniform grain scale. Documentation is silent about the latter.
A bug? Limitation? Problem with my OpenCL libraries?
Note: my OpenCL device is set to GPU.
