Enable Real-Time AI for High-Speed Data Acquisition with DAQIRI

2026-06-22 · Source: NVIDIA Technical Blog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Robotics & Autonomous Systems · Depth: Advanced, medium

Summary

NVIDIA DAQIRI (Data Acquisition for Integrated Real-time Instruments) is a high-performance networking library, part of the NVIDIA Holoscan Platform, designed to enable real-time AI for high-speed data acquisition. It shifts from inflexible hardware-centric designs to an adaptable, software-centric architecture, directly connecting high-bandwidth streaming detectors and sensors to the NVIDIA software ecosystem. DAQIRI allows developers to create workflows for in-stream processing tasks such as filtering, inference, compression, and adaptive control without modifying instruments. This is critical for facilities like the Linac Coherent Light Source II (LCLS-II), generating 1 MHz photon pulses, and the High-Luminosity Large Hadron Collider (HL-LHC) at CERN, which will increase luminosity by a factor of 10. Under the hood, DAQIRI bypasses the Linux kernel using DPDK for zero-copy access, routing data directly from NIC to GPU DMA buffers at 100s of Gbps, ensuring low latency and high throughput. It provides C++ and Python APIs with YAML-driven configuration for ease of deployment, abstracting complex low-level networking.

Key takeaway

For AI Engineers and Research Scientists building real-time data acquisition systems, NVIDIA DAQIRI offers a critical shift from hardware-centric to software-defined processing. You should consider integrating DAQIRI to bypass traditional "store first" bottlenecks, enabling direct streaming of high-bandwidth sensor data to GPUs for immediate AI inference and adaptive control. This allows you to gain real-time insights and prepare high-quality, AI-ready data for downstream supercomputing analysis, significantly accelerating discovery.

Key insights

DAQIRI enables real-time AI processing of high-speed sensor data by directly streaming to GPUs, bypassing traditional bottlenecks.

Principles

• Data acquisition bottlenecks shift from missing data to "collect, store, analyze" architectures.
• Pre-processing data at the source mitigates information loss from data deluge.
• Kernel-bypass networking reduces latency and CPU overhead for high-speed data.

Method

DAQIRI uses DPDK for kernel-bypass, zero-copy data transfer from NIC to GPU DMA buffers, configured via YAML, and processed with C++/Python APIs.

In practice

• Stream instrument outputs directly into edge supercomputing systems.
• Configure data paths, NICs, GPU memory regions, and flow rules via YAML.
• Use C++ or Python APIs to receive GPU-ready tensors for inference.

Topics

• Real-Time AI
• Data Acquisition
• NVIDIA DAQIRI
• GPU Acceleration
• High-Bandwidth Networking
• Edge Computing
• DPDK

Code references

• NVIDIA/daqiri

Best for: AI Architect, MLOps Engineer, Machine Learning Engineer, AI Engineer, Research Scientist, Software Engineer

Related on AIssential

• The AI Industry Spent Billions Chasing Faster Chips. Inference Startup Kog Says They Were Solving the Wrong Problem. — Software optimization can enable standard GPUs to match specialized AI hardware performance for real-time inference.
• Add a Specialized Deep Research Skill to Agent Harnesses — NVIDIA AI-Q provides agent harnesses a dedicated, secure, and auditable deep research skill for enterprise data synthesis and analysis.
• The Hardware Bottleneck AI Can’t Fix — Hardware engineering's lack of software-like tooling creates a bottleneck, hindering AI's impact and requiring specialized data platforms.
• What AI benchmarks miss about real-world performance — Real-world AI performance is bottlenecked by fragile data paths between storage and compute, a factor often missed by standard benchmarks.
• Introducing AI Runtime: Scalable, Serverless NVIDIA GPUs on Databricks for Training and Finetuning — AI Runtime simplifies distributed GPU training by offering serverless, on-demand access and optimized tools within Databricks.
• AI's GPU problem is actually a data delivery problem — Inefficient data delivery, not GPUs, is the primary bottleneck for AI workload performance and scalability.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by NVIDIA Technical Blog.