I'm using ArcGIS Pro and Deep Learning API version 3.2.
I was told that this version now supports training deep learning models using a single machine with multiple GPUs. I have such a machine configured. I selected in the Environment panel GPU as processor type and left the GPU ID as blank since there are multiple GPUs.
Is this how to instruct ArcGIS Pro Deep Learning to use multiple GPUs? My model type is HED Edge Detector (Pixel Classification). Does anyone know if HED is supported for multiple GPUs? I have tried all possible model types and only HED came close to identifying field boundaries for my project.
Appreciate any response.
Hello @JadedEarth
By default, the tool uses all available GPUs when the Model Type parameter is set to one of the following
These are the only supported model types currently for multiple gpus. Also, the deep learning tools only support CUDA capable GPUs.
regards,
Pavan
That's unfortunate. I need Holistically-Nested Edge Detection (HED) model for my particular project. I've tried all the models you mentioned and none of them provided results like HED. I was hoping ESRI will support this model.
I have shared your feedback with the team.
thanks
Pavan
Dear @JadedEarth
The multi-gpu support for deep learning models listed here is available.
Please note: The multi-gpu training is supported with command line only. If you are trying to use ArcGIS Pro tools, it would not work.
Can you show me the link to instructions on how to run multi-gpu from command line? There should have been instructions/guide to do this from the outset. When you advertised that multi-gpu is supported, nowhere did it state that it only works for command line and there are no links to instructions on how to go about it.