In a modern Document Understanding scenario, what is the minimum required number of dedicated GPU-enabled agent servers?
In an enterprise Automation Suite deployment implementing modern Document Understanding projects without CPU fallback inference, UiPath hardware sizing standards require a minimum of five (5) dedicated GPU-enabled agent servers (or GPU allocations). Modern Document Understanding utilizes separate concurrent neural network pipelines to process unstructured business documents, invoices, and receipts. These five dedicated GPU instances are allocated across four functional roles: OCR processing pipelines, custom model training tasks, custom model inference pipelines, and out-of-the-box (OOTB) pre-trained model inference engines. Sharing a smaller number of GPUs across these distinct operations leads to GPU memory exhaustion (CUDA out-of-memory errors) and inference job queuing. Dedicating five GPU-enabled servers ensures each pipeline has the compute throughput required to sustain real-time document processing and training cycles. (Reference: UiPath Automation Suite Sizing Guide -- Modern Document Understanding GPU Sizing).