NVIDIA's PVA Performance Evaluation Binaries
The Programmable Vision Accelerator (PVA) is a low-power hardware accelerator available in Jetson Orin NX and AGX modules. Designed for vision processing tasks, it can significantly outperform traditional CPU-based implementations in both speed and power efficiency.
This free product provides a set of RidgeRun-developed PVA-optimized binary samples demonstrating key embedded vision and image signal processing (ISP) tasks. The goal is to showcase PVA capabilities and offer a performance benchmark against CPU equivalents. A detailed performance analysis can be found in our Developer's Wiki.
RidgeRun is a NVIDIA PVA Preferred Partner, with access to the complete toolchain required to design, develop, and deploy custom algorithms on the PVA, tools not widely accessible to the general public. We offer end-to-end support, from board bring-up to fully customized vision pipelines. Get in contact to discuss your specific needs.
Checkout the official documentation here: NVIDIA Programmable Vision Accelerator SDK.
PVA Key Capabilities:
- Offloading Pre-processing and Post-processing Tasks in AI Pipelines
- Executing Standalone Computer Vision Algorithms
- Performing Mathematical Computations
What’s Included:
These binaries are toy examples for you to get an idea of the PVA's performance boost. Binaries include CPU and PVA implementations of the following algorithms:
- Color Space Conversion: RGB to Gray
- Bit Shifting
- Lens Shading Correction
- 2D Convolution
Supported Platforms:
- NVIDIA Jetson AGX Orin
- NVIDIA Jetson Orin NX
Why RidgeRun?
With over 20 years of embedded software consulting experience, RidgeRun delivers performance-driven video processing tools tailored for real-time vision systems, embedded AI, and multimedia pipelines. These samples aim to to provide a performance reference for some PVA vs CPU algorithm implementations.
Learn more in our Developer's Wiki.