How to use GPU?

[wdk11]

I am using the latest version of gempyV3.0 and want to use GPU acceleration for modeling. I have successfully installed PyTorch, but when I run the command sol = gp.compute_model(geo_model,,
engine_config=gp.data.GemPyEngineConfig(
backend=gp.data.AvailableBackends.PYTORCH,
use_gpu=True,
dtype="float64"
))
I noticed that the GPU's video memory barely changed, and I was wondering if I was actually utilizing the GPU. How can I check if I'm using the GPU? Thank you.

[javoha]

Hi,

as far as I know the standard way to check if the torch configuration is working ist:

torch.cuda.is_available()

this should return True.

As a side note: We are currently working on resolving some overhead calculations in the core of gempy. It might be that we will be able to increase speed a little bit in the future.

Cheers.
Jan

[wdk11]

geo_model: gp.data.GeoModel = gp.create_geomodel(
project_name='Tutorial_ch1_1_Basics',
extent=[562298, 563540, 3032848, 3033348, 1800, 2700],
resolution=[50, 50, 50], # * Here we define the resolution of the voxels
# refinement=4, # * Here we define the number of octree levels. If octree levels are defined, the resolution is ignored.
importer_helper=gp.data.ImporterHelper(
path_to_orientations=data_path + "/orientations.csv",
path_to_surface_points=data_path + "/points.csv",
)
)
gp.map_stack_to_surfaces(
gempy_model=geo_model,
mapping_object= # TODO: This mapping I do not like it too much. We should be able to do it passing the data objects directly
{
"Strat_Series": ('dmx', 'Qdl', 'Qgl')
# "Strat_Series": ('dmx', 'Qdl')
}
)
gp.set_section_grid(
grid=geo_model.grid,
section_dict={
# 'section1':([563915.18,3033408.73],[562473.68,3031254.95],[500,500]
'section2':([563668.52,3033498.51],[562094.92,3032038.43],[500,500])
}
)
sol = gp.compute_model(geo_model,
engine_config=gp.data.GemPyEngineConfig(
backend=gp.data.AvailableBackends.PYTORCH,
use_gpu=True,
dtype="float32"
))
I have configured the GPU as required, and the result of torch.cuda.is_available() is True. Here is my code. My surface points number over 1000. My computer has 128GB of memory and an A5100 graphics card. After running, it tells me that there is insufficient memory, and there is no change in the GPU's video memory. I would like to ask why this order of magnitude of surface points can occupy 128GB of memory. Could you please tell me how to process the data or how to check if the GPU is being used,thank you.

[javoha]

Hi again,
so maybe to add a little context: Gempy uses the implicit universal co-kriging interpolator. Broadly speaking this is an interpolator that prefers sparse datasets. This of course depends on the size and complexity of your model, but 1000 surface data points and how many orientation (?) are quite a lot.
How many surfaces are you modeling and how complex is the geometry? Depending in that you can either downsample your data or at least start by modeling fewer surfaces to see if this works and looks correct.
If you are able to share your code and input data here, I can also try to run it and see if I have the same problem.
Cheers
Jan

[wdk11]

orientations.csv
points.csv

import numpy as np
import gempy as gp
import gempy_viewer as gpv

data_path = 'C:/Users/Administrator/Desktop'

geo_model: gp.data.GeoModel = gp.create_geomodel(
project_name='Tutorial_ch1_1_Basics',
extent=[562298, 563540, 3032848, 3033348, 1800, 2700],
resolution=[100, 100, 100], # * Here we define the resolution of the voxels

refinement=4, # * Here we define the number of octree levels. If octree levels are defined, the resolution is ignored.

importer_helper=gp.data.ImporterHelper(
path_to_orientations=data_path + "/orientations.csv",
path_to_surface_points=data_path + "/points.csv",
)
)
gp.map_stack_to_surfaces(
gempy_model=geo_model,
mapping_object= # TODO: This mapping I do not like it too much. We should be able to do it passing the data objects directly
{
"Strat_Series": ('dmx', 'Qdl', 'Qgl')

"Strat_Series": ('dmx', 'Qdl')

}
)
gp.set_section_grid(
grid=geo_model.grid,
section_dict={
'section1':([563915.18,3033408.73],[562473.68,3031254.95],[500,500]),
'section2':([563668.52,3033498.51],[562094.92,3032038.43],[500,500])
}
)
sol = gp.compute_model(geo_model,
engine_config=gp.data.GemPyEngineConfig(
backend=gp.data.AvailableBackends.PYTORCH,
use_gpu=True,
dtype="float64"
))

gpv.plot_2d(geo_model,section_names=['section1'])

Thank you.