Please provide complete information as applicable to your setup.
• Hardware Platform (Jetson / GPU)
amd64 cpu with a RTX 5060
• DeepStream Version
DeepStream 7.1
• TensorRT Version
10.12.0.36-1+cuda12.9
• NVIDIA GPU Driver Version (valid for GPU only)
Driver Version: 575.64
• Issue Type( questions, new requirements, bugs)
Bugs
• How to reproduce the issue ? (This is for bugs. Including which sample app is using, the configuration files content, the command line used and other details for reproducing)
Trying to run gst-launch-1.0 rtspsrc protocols=4 <…> latency=100 ! queue ! rtph264depay ! queue ! h264parse ! queue ! nvv4l2decoder ! queue ! .sink_0 nvstreammux batch-size=1 width=1920 height=1080 ! queue ! nvinfer config-file-path=<…>/dstest3_pgie_config.txt ! fakesink
Ive tried starting out with the triton development docker image on nvcr.io but it either ran the pipeline very slow(given that i piped the result to a filesink) or not at all
So i decided to run a custom docker built from 22.04 provided by canonical, and even though after fixing a bunch of issues, it ran on a 2060(with some issues but atleast it ran immediately with 4 streams and i) i cant get it to run without problems on a 5060.
dstest3_pgie_config.txt:
[property]
gpu-id=0
net-scale-factor=1
model-color-format=1
onnx-file=<…>YOLOX_Large_640x640_21_05_2025.onnx
model-engine-file=<…>/model_b1_gpu0_fp16.engine
labelfile-path=<…>/labels.txt
batch-size=1
network-mode=2
num-detected-classes=2
interval=0
gie-unique-id=1
process-mode=1
network-type=0
cluster-mode=2
maintain-aspect-ratio=1
symmetric-padding=0
parse-bbox-func-name=NvDsInferParseYoloCuda
custom-lib-path=/usr/local/lib/libnvdsinfer_custom_impl_Yolo.so
engine-create-func-name=NvDsInferYoloCudaEngineGet
[class-attrs-all]
nms-iou-threshold=0.45
pre-cluster-threshold=0.5
topk=300
the models are being built from onnx right as docker launches and baremetal ran perfectly on two different 5060 machines.
and as im writting this after a while of leaving the image(wrote this very post) to run the full program(that does the above pipeline and more) on its own, and out of nowhere it works, but seems to be lagging ever other second.
The program is a standalone binary that executes a pipeline a bit more complex than the above (in fact i launch 4 streams) but i noticed the main problem is between nvinfer and streammux/preprocessing as they tipically run beyond slow.
Replicated it by launching the docker(which on its own launches the program that does nothing as it waits for a connection) waited 10-20 minutes and suddenly the program works once again although lagged.
The graphics card seems under stress which is unusual as when i run the program outside of docker it runs rather smoothly although the first ever launch on a machine always goes wrong for some reason
• Requirement details( This is for new requirement. Including the module name-for which plugin or for which sample application, the function description)
Nvidia container toolkit is at 1.17.8-1