A developer has identified a consistent 43ms latency overhead in the Nvidia Argus library on Jetson Orin NX hardware, independent of ISP usage. The findings suggest a potential architectural bottleneck in the capture pipeline that impacts real-time performance for edge computing applications.
The identification of a fixed 43ms overhead in the Argus capture pipeline presents a significant hurdle for low-latency streaming and edge computing applications. Because this delay occurs independently of ISP usage, developers cannot simply bypass processing stages to achieve faster frame delivery on Jetson Orin NX hardware. This bottleneck forces a trade-off between using Nvidia's optimized library and seeking alternative, lower-level capture methods that may lack Argus's feature set. Industry observers should monitor Nvidia's developer forums for a formal response or kernel-level patch to address this architectural constraint in future JetPack releases.
The 43ms overhead reported in this thread is consistent with a broader pattern of Argus latency complaints on Jetson Orin NX hardware. In a separate forum thread from May 2025, a developer building a stereo visual inertial odometry system with externally triggered Sony IMX296 global shutter sensors measured over 30ms delay from trigger signal to frame availability using libargus on Jetson Orin NX, with latency scaling by resolution: 35ms at 1456x1088, 26-30ms at 1280x720, and 10-15ms at 728x544. Nvidia forum moderator ShaneCCC confirmed in that thread that frame rate is the key determinant of latency and recommended V4L2 as an alternative for latency-critical applications, noting that Argus has many internal pipeline stages that impact delay.
Nvidia has officially acknowledged that the Argus pipeline introduces a fixed multi-frame latency by design. In a March 2025 thread involving a GMSL external-triggered camera on Orin NX measuring approximately 370ms from sensor timestamp to user-space frame delivery, Nvidia staff confirmed that 4 to 6 frames of glass-to-glass latency is expected and represents the best achievable result for the current Argus design, with no supported method to reduce it while remaining within the Argus API. The only recommended path for lower latency is bypassing Argus entirely and using V4L2, which sacrifices access to the Jetson ISP's 3A algorithms.
Beyond latency, the Argus library on Jetson Orin NX has exhibited stability issues that compound the usability problem. Developers running multi-camera Argus capture sessions on Orin NX with JetPack 6.2.2 have reported frame capture abnormalities when combining specific camera index pairs, with certain configurations causing runtime failures while others work correctly on the same hardware. Earlier JetPack 5.x releases also required community-sourced patches to resolve Argus process hangs during long-running capture sessions, with Nvidia providing stability fixes for R35.6.x that improved but did not fully eliminate crashes at higher frame rates.
Developers have identified a consistent 43ms latency overhead within the Nvidia Argus library on Jetson Orin NX hardware. This delay persists regardless of ISP usage, effectively doubling glass-to-glass latency. This bottleneck forces developers to choose between using the Argus library or switching to V4L2 for lower-latency, performance-critical applications.
Benchmarks on the Jetson Orin NX hardware have identified a consistent 43ms 'ExtraTime' overhead within the Argus capture pipeline.
No, the 43ms latency overhead remains identical whether the ISP is active or bypassed, suggesting an architectural bottleneck in the capture pipeline.
Nvidia staff and forum moderators recommend using V4L2 as an alternative to Argus for latency-critical applications, though this sacrifices access to the Jetson ISP's 3A algorithms.
Yes, Nvidia has confirmed that the Argus pipeline is designed with multi-frame latency, noting that 4 to 6 frames of glass-to-glass delay is expected behavior.
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source