Show HN: We built an OCR server that can process 270 dense images/s on a 5090

Category: infrastructure

Tags: ocr, gpu-accelerated, cuda, tensorrt, document-processing

Score: 8.0/10 (Innovation: 7, Technical: 9, Documentation: 9, Utility: 7)

TurboOCR is a high-performance GPU-accelerated OCR server written in C++/CUDA/TensorRT, achieving 270 images per second on dense documents with latency under 11ms. It combines multiple innovative approaches including native PDF handling with security-aware mode selection, layout detection, and a dual HTTP/gRPC API, making it significantly faster than alternatives like PaddleOCR while maintaining competitive accuracy.

Target audience: backend devs, data engineers, devops

Repository: https://github.com/aiptimizer/TurboOCR · C++ · MIT · 235 stars

View on Hacker News