Allocating GPU resources for Document Understanding modern projects

Allocate GPU resources to Document Understanding services through ArgoCD parameters on EKS/AKS.

For optimal performance, allocating the appropriate number of GPU resources is vital. You allocate GPU resources through ArgoCD, which contains specific parameters that correspond to each service. By adjusting these parameters, you can control the distribution of GPU resources as per the needs of each service.

Each service listed in the capacity calculator has a corresponding ArgoCD parameter as follows:

• OCR replicas corresponds to the du-ocr-saas.replicaCount parameter.
• Custom model training replicas corresponds to the du-training-ssde.replicas parameter.
• Custom model replicas corresponds to the du-aimodelhost-2404.replicas parameter.
• OOB model replicas corresponds to the du-ssde-saas.replicaCount parameter.

Before allocating of GPU resources to each service, use the capacity calculator to determine the appropriate number of GPUs that should be allocated to each service. The output is based on your specific needs and will serve as a guide when updating the parameters in ArgoCD.

For details, refer to the UiPath Automation Suite Install Sizing Calculator page.

Note:

For Document Understanding modern projects, the minimum recommended GPU is NVIDIA T4.

1. Open ArgoCD.

2. Select the documentunderstanding application.

3. Select Details to open the Details menu.

   

4. Select the Parameters tab.

5. Select Edit so you can edit the parameter values.

   

6. Search for the parameters that need to be edited and allocate the needed number of GPUs.

   Each service listed in the capacity calculator has its corresponding ArgoCD parameter as follows:

   • OCR replicas corresponds to the du-ocr-saas.replicaCount parameter.
   • Custom model training replicas corresponds to the du-training-ssde.replicas parameter.
   • Custom model replicas corresponds to the du-aimodelhost-2404.replicas parameter.
   • OOB model replicas corresponds to the du-ssde-saas.replicaCount parameter.

7. Select Save.