Orply.

DAQIRI Streams Berkeley Lab Camera Data at More Than 500 Gbps

NVIDIAMonday, November 16, 20263 min read

NVIDIA says its DAQIRI system streams high-bandwidth instrument data into GPU memory for analysis while an experiment or industrial process is still underway, rather than waiting for data to be written to disk. At Berkeley Lab’s Molecular Foundry, researchers use it to process more than 500 gigabits per second from a new camera and receive results during the experiment. The company argues that live analysis can help guide the next measurement or response.

The camera example makes real-time processing concrete

At Berkeley Lab’s Molecular Foundry, researchers have faced a specific imaging problem: seeing strong and faint signals together without losing fine detail. A new camera captures both, and NVIDIA DAQIRI streams its output directly into GPU memory for processing. The results arrive during the experiment, not after the data has been written to disk and analyzed.

500+ Gbps
camera data streamed directly into GPU memory

The comparison shown alongside the example puts the scale in context. A Dual EELS measurement is labeled at less than 1 Gbps; the DoEELS camera is labeled at more than 500 Gbps, with frames being summed. The contrast is not simply between a slower and a faster camera. It shows the volume of data that the workflow must handle while the experiment is still in progress.

DAQIRI’s role is to connect that high-bandwidth stream to accelerated computing for immediate processing. That timing matters because results can still inform the ongoing experiment. The camera captures the signals; the live stream makes processing possible before the measurement is over.

Live analysis can change what an instrument does next

In a conventional pipeline, an instrument’s data passes through memory, is written to disk, and is analyzed later. That sequence separates measurement from interpretation. For a fleeting scientific event or an urgent industrial decision, the delay can mean that the result arrives too late to affect the work underway.

DAQIRI is presented as an alternative: instrument output is made available to GPU memory and inference as it arrives. The diagrams illustrate two possible uses. One searches a live stream for a specific event and marks when it is found. Another processes a full-field scan and returns an update that directs a focused scan. In the latter case, analysis can shape the next measurement rather than merely describe a completed one.

The broader claim is that accelerated AI can process sensor data while an experiment or industrial process is still underway. The workflow can then sense, reason, and respond in real time. That does not mean every instrument automatically acts on every result; the examples show a path from live data, through processing, to information that can guide a next step.

The same pattern reaches beyond laboratory imaging

NVIDIA identifies several other applications for live instrument data: finding promising areas within a scan, characterizing a telescope signal, guiding the next step of an experiment, and keeping interactive digital twins synchronized with complex physical systems. Each depends on getting useful results while the instrument or physical system is still operating.

The settings shown range from radio telescopes and industrial testing to fusion and automated laboratory instrumentation. The examples broaden the proposed use of DAQIRI, but the camera at Berkeley Lab is the one tied to a specific data rate and to results delivered during an experiment.

The dividing line is therefore not simply whether a system can collect large volumes of data. It is whether that data can be processed soon enough to matter to the process producing it. DAQIRI’s proposition is to put high-bandwidth instrument streams and accelerated computing together within that working window.

The frontier, in your inbox tomorrow at 08:00.

Sign up free. Pick the industry Briefs you want. Tomorrow morning, they land. No credit card.

Sign up free