Using ESRI Deep Learning Tool to create a feat class

AI & Data Science Deep Learning (Training & Inference) JAX
1

Hi, I am trying to run a deep learning library and I am getting errors involving CUDA. My GPU is a NVIDIA RTX A1000 Laptop GPU and the current driver is 31.0.15.1801.
I just want confirmation whether my gpu should be able to run CUDA? If so, what do you think the solution is here?

Detect Objects Using Deep Learning

Tool Path

Input Raster hxip_m_3712102_ne4_10_15.tif
Output Detected Objects C:\Users\carmen.gonsalves\OneDrive - California Department of Technology\Documents\ArcGIS\Projects\Deep_Learning_Bldg_FtPnts\Deep_Learning_Bldg_FtPnts.gdb\sd73a73c9_450f_4f20_b0d6_26de7e0c2e23
Model Definition C:\Users\carmen.gonsalves\OneDrive - California Department of Technology\Documents\ArcGIS\Projects\Deep_Learning_Bldg_FtPnts\usa_building_footprints.dlpk
Arguments padding 128;batch_size 4;threshold 0.9;return_bboxes False;test_time_augmentation False;merge_policy mean;tile_size 512
Non Maximum Suppression NMS
Confidence Score Field Confidence
Class Value Field Class
Max Overlap Ratio 0
Processing Mode PROCESS_AS_MOSAICKED_IMAGE
Output Classified Raster
Use pixel space NO_PIXELSPACE
Objects of Interest

Environments

GPU ID 1
Extent 608138.757198425 4200447.78595189 609799.720809342 4201469.73995138 PROJCS[“NAD_1983_2011_UTM_Zone_10N”,GEOGCS[“GCS_NAD_1983_2011”,DATUM[“D_NAD_1983_2011”,SPHEROID[“GRS_1980”,6378137.0,298.257222101]],PRIMEM[“Greenwich”,0.0],UNIT[“Degree”,0.0174532925199433]],PROJECTION[“Transverse_Mercator”],PARAMETER[“False_Easting”,500000.0],PARAMETER[“False_Northing”,0.0],PARAMETER[“Central_Meridian”,-123.0],PARAMETER[“Scale_Factor”,0.9996],PARAMETER[“Latitude_Of_Origin”,0.0],UNIT[“Meter”,1.0]]
Processor Type GPU

Messages

Start Time: Monday, October 6, 2025 3:43:48 PM
Working with cell size 0.150000 (less than 1.00) meter.
ERROR 999999: Something unexpected caused the tool to fail. Contact Esri Technical Support (http://esriurl.com/support) to Report a Bug, and refer to the error help for potential solutions or workarounds.
Unspecified error [Failed to generate table]
[Missing raster: [%s]]
Unable to initialize python raster function with scalar arguments. [C:\Users\CARMEN~2.GON\AppData\Local\Temp\ArcGISProTemp2820\usa_building_footprints.dlpk\ArcGISInstanceDetector.py]
Traceback (most recent call last):
File “C:\Users\CARMEN~2.GON\AppData\Local\Temp\ArcGISProTemp2820\usa_building_footprints.dlpk\ArcGISInstanceDetector.py”, line 162, in initialize
torch.cuda.set_device(device)
File “C:\Program Files\ArcGIS\Pro\bin\Python\envs\arcgispro-py3\Lib\site-packages\fastai\torch_core.py”, line 72, in _new_torch_cuda_set_device
_old_torch_cuda_set_device(device)
File “C:\Program Files\ArcGIS\Pro\bin\Python\envs\arcgispro-py3\Lib\site-packages\torch\cuda\_init_.py”, line 350, in set_device
torch._C._cuda_setDevice(device)
RuntimeError: CUDA error: invalid device ordinal
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.
Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.
Unable to initialize python raster function with scalar arguments.
Failed to execute (DetectObjectsUsingDeepLearning).
Failed at Monday, October 6, 2025 3:44:27 PM (Elapsed Time: 38.23 seconds)

2

Hi Carmen,

Your GPU is compatible with CUDA per this post here!

NVIDIA RTX A1000 Laptop GPU cuda compatibility

Take a look at the CUDA installation guides:

And in the meantime, I will slack the team and see if someone has the bandwidth to take a look!

Thanks,

AHarpster

3

I work with Carmen and am also debugging this. Two things to add.

First, can you speak to why the A1000 isn’t listed here? CUDA GPU Compute Capability | NVIDIA Developer
I do see CUDA listed in the device specs though, but on a third party site: NVIDIA RTX A1000 Mobile Specs | TechPowerUp GPU Database

Second, the application in question is ArcGIS Pro. Their developers provide deep learning/CUDA libraries as an installation package on top of their main installation. That installer and details are here: GitHub - Esri/deep-learning-frameworks: Installation support for Deep Learning Frameworks for the ArcGIS System

My understanding is that PyTorch in this instance should be installed as a conda package, so the binaries should be included and compatible with each other. From reading online, my undersanding is that if the conda package of pytorch is installed, we shouldn’t need to run the CUDA toolkit installer (it won’t use it), just the driver install, but I’d love to know if this is correct or incorrect.

Carmen was originally getting an error about the driver version being older than the CUDA version, but we upgraded the driver and now she receives what she posted. We installed the latest driver (from July) from HP for our model of computer, but are going to test installation using NVIDIA’s driver package next in case HP’s driver distribution is missing something for CUDA.

Thank you!

4

@Aharpster Any further movement on this by chance?