YantraVision’s AMD NPU
Services
AMD-NPU | Deeplearning | Image and Signal Processing
AMD Ryzen AI processors integrate an on-die NPU based on the XDNA architecture to execute AI workloads locally, independent of CPU and GPU resources.
This enables low-latency, power-efficient inference for vision, media, and assistant workloads across Ryzen AI platforms, reaching throughput of ~50–60 TOPS.
Why work with YantraVision?
YantraVision provides end-to-end engineering for AMD Ryzen AI NPUs, covering system design, model optimization, and deployment using the Ryzen AI software and AI Engine stack.
We support teams from prototype to production, enabling on-device AI workloads without in-house NPU expertise, and deliver optimized, production-ready AI features on Ryzen AI platforms.
Our Core offerings
NPU architecture and feasibility :
- Analyze workloads across CPU, integrated GPU, and NPU.
- Partition execution using Ryzen AI software and ONNX Runtime.
- Define model selection, quantization approach, and performance targets per Ryzen AI generation.
Custom NPU Applications :
- Develop custom AI Engine (AIE) kernels using MLIR-AIE.
- Design multi-tile graphs with DMA and streaming dataflow.
- Implement image processing, signal processing, and pre/post-processing pipelines.
Validation, tooling, and enablement :
- Benchmark execution across CPU, GPU, and NPU
- Perform regression and long-run stability testing on Ryzen AI platforms
- Deliver documentation, reference implementations, and developer training
Overview
We work in the complete spectrum from system architecture to product realization with expertise in
Key Highlights
- System Design
- HDL’s / Higher Level Synthesis
- Custom IP Development
- 3rd Party IP Integration
Tools
- FPGA Tools – Xilinx Vivado, HLS – Vivado
- Intel Quartus, Intel HLS
- Simulation – ModelSim, NCSIM
- Others – Xilinx SDSoC, Xilinx reVISION, Intel SoC EDS
Hardware Layer
Our expertise in FPGA Hardware includes RTL coding, Higher level Synthesis, Verification and Validation. Our System Design capabilities ensures efficient hardware partitioning, hardware-software co-development to keep Time to Market faster
Key Highlights
- MicroEngine, Hardware Kernel & DataFlow Computing
- Image, Video and Audio Processing
- ML and DL Optimised Implementation
- Algorithm Porting
Tools
- openCL
- xfOpenCV, xfDNN
- AWS-F1
OS / Firmware Layer
Our expertise in the FPGA Firmware layer includes device driver integration and kernel module development. Depending on the application we work on multiple abstractions like BareMetal & Linux Systems etc.
Key Highlights
- Kernel and Firmware development for BareMetal platforms
- Linux based platforms
- Device drivers, Kernel module development for IP’s
- High performance – zero copy – kernel modules
Tools
- Yocto
- Petalinux
- Xilinx SDK
- Intel SDK
Middleware Layer
In the middleware layer we have extensive experience in implementing codec like J2K and h264 / h265. We also have experience in device driver integration with the middleware stack.
Key Highlights
- J2K
- Multichannel H.264/H.265 Codec Integration
- Device Driver Integration
Tools
- GStreamer
- FFMPEG
Application Layer
We develop image processing and machine learning applications that can be ported on FPGA for edge applications. When we build applications on FPGA we take full advantage of its capabilities like multicore CPU, GPU, PL and DSP elements so that platform is highly optimised.
Key Highlights
- Image, Video and Audio Processing
- ML and DL Optimised Implementation
- Algorithm Porting
Tools
- Python Libraries
- C, C++ based Applications
- OpenCL, OpenCV
- OpenGL
Example Application Domains
Computer vision & imaging
- Real-time image conversions.
- Image enhancement, denoising, and background removal.
- Background Blur, noise suppression.
- On-device inspection and camera-based analytics.
- Smart Framing, low-latency inference.
- Classical vision pipelines on AIE tiles.
Industrial Applications
- Agri Processing Application: Sorting
- Textile Application: Fabric Inspection
- Pharma Application: Tablet Inspection
- Print Application: Print Quality Inspection