Twenty-eight pages per second from a mid-range card — that translates to roughly 2.4 million scanned pages per day if you let it run flat out. We tested PaddlePaddle PP-OCRv4 (detection + recognition pipeline) on the NVIDIA RTX 5060 (8 GB VRAM) using a GigaGPU dedicated GPU server, and it hits a compelling balance of speed, cost, and simplicity for document-heavy operations.
Throughput Metrics
| Metric | Value |
|---|---|
| Pages/sec | 28 pages/sec |
| Latency per page | 35.7 ms |
| Precision | FP16 |
| Pipeline | Det + Rec + Cls |
| Performance rating | Good |
Benchmark conditions: FP16 inference, batch size 1, PP-OCRv4 full pipeline (detection + direction + recognition) on A4-format document scans.
Memory Profile
| Component | VRAM |
|---|---|
| Model weights (FP16) | 1.2 GB |
| Processing buffer | ~0.4 GB |
| Total RTX 5060 VRAM | 8 GB |
| Free headroom | ~6.8 GB |
PaddleOCR only consumes about 1.6 GB total, leaving nearly 7 GB available. That spare capacity is practical: you could run a quantised 7B LLM alongside OCR for immediate entity extraction, or increase the batch size for even higher throughput on multi-page documents.
Cost per Page
| Cost Metric | Value |
|---|---|
| Server cost | £0.35/hr (£99/mo) |
| Cost per 1M pages | £3.47 |
| Pages per £1 | 288184 |
At £3.47 per million pages, the RTX 5060 delivers the best cost-per-page ratio in our PaddleOCR lineup. The card itself is inexpensive, the throughput is solid, and PaddleOCR’s slim footprint means the GPU is never bottlenecked on memory. See all benchmarks for a full cross-GPU comparison.
Best Use Cases
The 5060 is a strong pick for small-to-medium document processing services: receipt scanning, invoice digitisation, form data extraction, or building searchable PDF archives. If you are running a PaddleOCR hosting setup and your throughput needs sit comfortably under 100,000 pages per hour, this card handles it without breaking a sweat — and without breaking the budget.
Quick deploy:
docker run --gpus all -p 8866:8866 paddlecloud/paddleocr:latestSee our PaddleOCR hosting guide, best GPU for OCR, and all benchmark results. Related: LLaMA 3 8B on RTX 5060 benchmark.
Deploy PaddleOCR on RTX 5060
Order this exact configuration. UK datacenter, full root access.
Order RTX 5060 Server
