Hardware Requirements

Running the Document Understanding ML Packages on a GPU includes an optimization meant to accelerate the training process.

Compatibility Matrix

As a result, training on GPU is five times faster than on CPU (previously it was 10-20 times faster). This also makes it possible to train models on CPU with up to 5000 pages (previously it was 500 maximum).

Please be aware that training Document Understanding models on GPU requires a GPU with at least 11GB of video RAM to run successfully.

Use the below table to check the compatibility between the ML Packages, CUDA version, and GPU driver version.

ML Packages version	CUDA version	NVIDIA driver (lowest compatible version)	Hardware Generation
2022.10	CIDA 11.3	R450.80.03	Ampere, Turing, Volta, Pascal, Maxwell, Kepler
2022.4	CUDA 11.3	R450.36.06	Ampere, Turing, Volta, Pascal, Maxwell, Kepler

CUDA is backward compatible, meaning that existing CUDA applications can continue to be used with newer CUDA versions.

More information about compatibility can be found here.

CPU and GPU Usage

You can use the Document Understanding framework for reading text using an OCR engine, classifying the documents, and extracting information from the documents. While classification and extraction tasks are run on CPU, the OCR is recommended to be run on GPU (through a CPU version is also provided in case a GPU is not available).

The On-premises deployment is done using Automation Suite and its hardware requirements.

You can use the same type of VM for both extractors and classifiers, the only difference being the infrastructure size. We recommend using the OCR engine with a GPU VM. The compatibility between the ML Packages, CUDA version, and GPU driver version are described in the Compatibility Matrix section.

Let's take an actual example for better understanding the hardware requirements.

Service	Hardware requirement	Capability
Extractor	Use a VM with minimum 2 CPU cores and 8 GB RAM	Can process 25,000 pages/day or 5 million pages/year, assuming perfectly constant traffic (no spikes).
Classifier	Use a VM with minimum 2 CPU cores and 8 GB RAM	Can process 40,000 documents/day or 8 million documents/year, assuming perfectly constant traffic (no spikes).
OCR	Recommended to be used with GPU VM	Can process 50,000 pages/day.

Example : If you process 10 million pages/year, then you need a VM with 4 CPU cores, 16 GB RAM for the extractor, another one for the classifier, and a third VM with an NVidia GPU core for the OCR engine.

You can also choose to use only one VM for both extractor and classifier, meaning that you need a single VM with 8 CPU cores and 32 GB RAM.

Note: You can always use more more powerful CPU/GPU VMs for increasing the number of processed documents/day.