RidgeRun Service: Port or Develop Custom Computer Vision Algorithms for NVIDIA's PVA to Maximize Performance
- Marco Herrera
- Jun 19
- 4 min read
TL;DR: RidgeRun now has the tools and expertise to create, port, and optimize algorithms directly for NVIDIA’s PVA; a powerful accelerator for vision tasks on Jetson platforms. Algorithms offloaded to the PVA run faster and more efficiently than on the CPU, freeing up system resources and reducing power consumption. If you’re looking to improve performance in real-time vision or signal processing applications, RidgeRun can help you implement and accelerate your algorithms on the PVA, from initial feasibility to full deployment. Contact us.
At RidgeRun, we seek new opportunities and emerging technologies to enhance our clients' products. One of the most exciting developments we’re currently exploring is NVIDIA’s PVA programming, which enables developers to take full advantage of the Programmable Vision Accelerator (PVA) for offloading and accelerating vision-specific tasks.
While the GPU tends to get most of the attention, it is not the only co-processing unit available on NVIDIA Jetson platforms. Jetsons also include other specialized hardware accelerators like the Video Image Compositor (VIC), the NVIDIA Deep Learning Accelerator (NVDLA), and the PVA. These accelerators reduce the system's dependence on the CPU and GPU, making it possible to design more efficient and responsive AI and computer vision applications. The PVA, in particular, opens up new possibilities for real-time vision processing with lower power consumption and resource usage.
It is especially well-suited for tasks such as:
Low-power image signal processing for drone’s on-board computing.
Low-latency robotics signal and data processing: LiDaR, video, IMU motion data.
NVIDIA Holoscan operators accelerated by PVA.
Accelerated GStreamer real-time video processing.
The PVA is a fully programmable, multi-core VLIW SIMD vector processor specifically designed for efficient execution of computer vision tasks. It is ideal for offloading and accelerating compute-heavy image and video processing tasks from the CPU and GPU, enabling more efficient system performance. Find an overview of the architecture in the image below:
This is especially valuable for applications like computer vision, real-time data processing, deep learning inference, and high-throughput signal processing, scenarios where traditional processing units can quickly become overloaded.
By shifting parts of the processing pipeline to the PVA, developers can:
Free up CPU and GPU resources for other tasks.
Reduce power consumption and thermal load.
Improve overall system throughput.
Achieve faster, more efficient performance on embedded systems.
Leveraging the PVA isn’t just a performance booster; it’s a strategic tool that can help you scale smarter, especially in scenarios where efficiency and speed are critical. As workloads grow in complexity, having the option to delegate tasks to a dedicated accelerator is becoming less of a luxury and more of a necessity.
We've already explored and implemented a range of algorithms using the PVA, and the initial performance results are promising. Early tests show significant potential for offloading compute-heavy tasks, with noticeable efficiency and resource usage improvements. The plot from Figure 2. shows a performance comparison between CPU and PVA for a Radial Lens Shading Correction algorithm at different resolutions.
Check out more performance comparisons and general information about RidgeRun’s PVA expertise in our RidgeRun NVIDIA PVA Development developers wiki. Additionally, you can get access to the toy examples' binaries for our PVA implementation of Color Space Conversion, Radial Lens Shading Correction, Bit Shifting and 2D Convolutions in our shop.
As an NVIDIA Preferred Partner, RidgeRun is uniquely positioned to support teams exploring the potential of the PVA. Whether you're evaluating feasibility, looking to port existing code, or developing new algorithms from the ground up, our hands-on experience with the full PVA development flow—setup, integration, debugging, and profiling—can significantly shorten your path to success. We understand the tools, the challenges, and the opportunities, and we’re ready to help you accelerate adoption while reducing both risk and development time.
Our R&D team is actively working with this acceleration device and experimenting with migrating existing algorithms to exploit its capabilities thoroughly. Our current efforts are focused on:
General Image Signal Processing
Live Electronic-Assisted Video Stabilization
Advanced 360° camera stitching
Birds-Eye View
What's next?
If you're considering whether this technology is a good fit for your product, we’d love to help you explore the possibilities. Our team can assist at any stage—whether you're starting from an existing algorithm or building something entirely new.
Check out the diagram below to get a clearer picture of our PVA development workflow. It outlines how RidgeRun approaches each project: from evaluating and porting existing code or designing custom algorithms, to profiling performance, optimizing for the PVA, and delivering a fully integrated solution tailored to your system’s needs.
If you have questions or want to dive deeper into the PVA, we're here to help. RidgeRun can provide guidance, share insights from our ongoing work, and help assess the best path forward for your specific use case. Contact us, we’re always happy to talk, collaborate, and explore how this technology can bring real value to your system.
