Live-RTSP-only frame corruption on Jetson AGX Thor / DeepStream 9.1 — root cause + confirmed fix (uridecodebin vs explicit rtspsrc chain)

Hardware: Jetson AGX Thor
JetPack / L4T: 7.2 / R39.2
DeepStream: 9.1
GStreamer: 1.24.2
Cameras: RTSP IP cameras, H.264, 2688x1520 (4MP), 20fps, ~6 Mbps

Symptom — Dual-camera DeepStream pipeline shows intermittent, sudden blocky/mosaic corruption in decoded frames, live cameras only, never reproduced on file-replay of identical recorded footage. GStreamer’s bus reports zero errors/warnings during confirmed corruption. Episodes aren’t single-frame — a logged episode on 2026-08-14 shows continuous corruption sustained for ~8 minutes with a matching fps drop.

Evidence — instrumented the unfixed uridecodebin baseline with NVIDIA’s nvds_logger, a GStreamer bus watch, a 10s heartbeat, and tegrastats hardware telemetry. Captured a full episode (12:49–12:57 IST) gap-free: fps drops from steady ~4-5fps to erratic 1.4-2.4fps during corruption and recovers after; zero bus ERROR/WARNING/EOS the entire time despite the watch being confirmed active; no thermal/memory correlation.

Root cause — uridecodebin’s auto-negotiated internal pipeline for rtsp:// sources.

Additional layer — NVIDIA’s own unapplied update_rtpmanager.sh (4 patches), including a real deadlock fix and a packaging gap where one patch is missing from both the SDK install and the release tarball.

What did NOT fix it — nvurisrcbin, CPU decode (works but not viable), VIC exhaustion (ruled out), decoder property tweaks, and — notably — raising rtspsrc latency made it worse.

Open questions:

  1. Can anyone confirm uridecodebin as root cause on AGX Thor specifically?
  2. Is the missing rtpmanager patch (#2, the deadlock fix) intentionally excluded from the DS9.1 tarball, or an oversight?
  3. Guidance on backporting DS9.1-only patches to DS8.0 without a JetPack upgrade?
  4. Is the fps-drop signature during corruption a known symptom, or a separate contributing factor (jitterbuffer under packet loss)
  5. nvurisrcbin, did not fix this corruption on two separate Jetson AGX Thor devices, even though it’s the officially recommended RTSP source element for DeepStream. Is this a known limitation of nvurisrcbin, and if so, is manually constructing the RTSP pipeline from individual elements (rtspsrc, rtph264depay, h264parse, nvv4l2decoder) instead of using a bundled source-bin element the currently recommended approach for live RTSP on this hardware/DeepStream version?

Please refer these logs and the NVIDIA forum threads reviewed + what we tried from each:

02_corruption_episode_1249_1258.log (38.7 KB)

01_full_run_startup_to_1238.log (177.8 KB)

FORUM_LINKS_AND_FIXES_TRIED.txt (7.8 KB)

This forum is exclusively for developers who are part of the NVIDIA DRIVE® AGX SDK Developer Program | NVIDIA Developer . To post in the forum, please use an account associated with your corporate or university email address. This helps us ensure that the forum remains a platform for verified members of the developer program.

I have a corporate email address, haripratap@urmedz.in, but I am not receiving the OTP when trying to log in. I have also checked my spam/junk folder, but I still haven’t received the login OTP. I have tried multiple times, but the OTP is not coming through.

My personal email address is haripratapsn@gmail.com

This forum is for DRIVE Thor platform, you should post your DeepStream issue at Latest Intelligent Video Analytics/DeepStream SDK topics - NVIDIA Developer Forums