Consolidated FPS is limited to 100 Fps, irrespective of number of sources

  • GPU Nvidia A10
  • DeepStream 6.3
  • NVIDIA GPU Driver Version 535.154.05
  • Issue Type - Question/Bug


  1. We are running both pgie and sgie, pgie is yolov4 model with 1888x1056, sgie is a custom classifier.
  2. Running over 40 cameras live feed.
  3. At 1080p resolution, Every camera is expected to run at 5 fps.
  4. All cameras are on H264


  1. When running 40+ cams with expected 5 Fps, we ideally need to get consolidated 200+ Fps, but we are getting only fixed 100 Fps.
  2. Tested same setup with few senarios
    1. 18 cams, ideal Fps is 90, we got 90 Fps.
    2. 20 cams, ideal Fps is 100, we got 100 Fps.
    3. 24 cams, ideal Fps is 120, we still got only 100 Fps.
    4. irrespective of number of cameras, we are getting same 100 Fps.
  3. with and without classifier, the Fps remained same.


  1. What can be the choke/limiting point here?
  2. is there any limitations to number frames/sources can be decoded ?
  3. What can be done to achieve higher Fps?

Added some code similar to ours

main pipeline creation code

Standard GStreamer initialisation


# Create gstreamer elements
pipeline = create_pipeline()

# Create nvstreammux instance to form batches from one or more sources.
streammux = create_streammux()

streamdemux = create_streamdemux()

for source_num in len(sources):
    source = sources[source_num]
    source_bin = create_source_bin(source_num, source, drop_frame_rate)
    if not source_bin:
        sys.stderr.write("Unable to create source bin \n")
    padname = "sink_%u" % source_num
    sinkpad = streammux.get_request_pad(padname)
    if not sinkpad:
        sys.stderr.write(f"Unable to create sink pad bin for source {source_num} \n")
    srcpad = source_bin.get_static_pad("src")
    if not srcpad:
        sys.stderr.write("Unable to create src pad bin \n")

pgie = create_pgie()

# create tracker
if config.use_nvtracker:
    tracker = create_tracker([source_num for source_num in len(sources)])

# Use convertor to convert from NV12 to RGBA as required by nvosd
nvvidconv = create_nvvidconv([source_num for source_num in len(sources)])

# Create OSD to draw on the converted RGBA buffer.
nvosd = create_nvosd([source_num for source_num in len(sources)])
# nvvidconv_postosd = create_nvvidconv()

RTSP Out Code Start
# Create a caps filter
caps = create_capsfilter([source_num for source_num in len(sources)])

# Make the encoder
encoder = create_encoder([source_num for source_num in len(sources)])

# Make the payload-encode video into RTP packets
rtppay = create_rtppay([source_num for source_num in len(sources)])

# Make the UDP sink
udpsink_start_port = 5400
udpsink_port_list = [port for port in range(udpsink_start_port, udpsink_start_port+len([source_num for source_num in len(sources)]))]
sink = create_udpsink([source_num for source_num in len(sources)], udpsink_port_list)

RTSP Out Code End
streammux = add_streammux_props(streammux, frame_width, frame_height)  # setting properties of streammux
pgie.set_property('config-file-path', primary_model_config_file)  # setting properties of pgie


if config.use_nvtracker:
    dum_var = [pipeline.add(v) for v in tracker.values()]
dum_var = [pipeline.add(v) for v in nvvidconv.values()]
dum_var = [pipeline.add(v) for v in nvosd.values()]
dum_var = [pipeline.add(v) for v in caps.values()]
dum_var = [pipeline.add(v) for v in encoder.values()]
dum_var = [pipeline.add(v) for v in rtppay.values()]
dum_var = [pipeline.add(v) for v in sink.values()]

# Link the elements together:
# uridecodebin -> streammux -> nvinfer ->
# nvtracker -> nvvidconv -> nvosd -> nvvidconv_postosd ->
# caps -> encoder -> rtppay -> udpsink

for source_num in len(sources):
    source = sources[source_num]
    srcpad1 = streamdemux.get_request_pad(f"src_{source_num}")
    if not srcpad1:
        sys.stderr.write(" Unable to get the src pad of streamdemux \n")
    if config.use_nvtracker:
        sinkpad1 = tracker[f"tracker_{source_num}"].get_static_pad("sink")
        if not sinkpad1:
            sys.stderr.write(" Unable to get sink pad of tracker \n")
        sinkpad1 = nvvidconv[f"nvvidconv_{source_num}"].get_static_pad("sink")
        if not sinkpad1:
            sys.stderr.write(" Unable to get sink pad of nvvidconv \n")
    if config.use_nvtracker:

# create an event loop and feed gstreamer bus mesages to it
bus = pipeline.get_bus()
bus.connect("message", bus_call, loop)

# Start streaming
rtsp_port_num = 8554

server =
server.props.service = "%d" % rtsp_port_num

factory_dict = dict()
for i in range(len(sources)):
    udpsink_port = udpsink_port_list[i]
    factory =
        "( udpsrc name=pay0 port=%d buffer-size=524288 caps=\"application/x-rtp, media=video, clock-rate=90000, encoding-name=(string)%s, payload=96 \" )" % (
            udpsink_port, codec))
    factory_dict[f"factory_{i}"] = factory
for source_num in len(sources):
    server.get_mount_points().add_factory("/ds-test", factory_dict[f"factory_{source_num}"])

osdsinkpad = dict()
for source_num in len(sources)
    osdsinkpad[f"osdsinkpad_{source_num}"] = nvosd[f"nvosd_{source_num}"].get_static_pad("sink")
    if not osdsinkpad[f"osdsinkpad_{source_num}"]:
        sys.stderr.write(" Unable to get sink pad of nvosd \n")

for osp, source in zip(osdsinkpad.values(), sources):
    osp.add_probe(Gst.PadProbeType.BUFFER, osd_sink_pad_buffer_probe, 0, source)

# start play back and listen to events
print("Starting pipeline \n", pipeline)

Creating Source

def cb_newpad(decodebin, decoder_src_pad, data):

    print("In cb_newpad\n")
    caps = decoder_src_pad.get_current_caps()
    gststruct = caps.get_structure(0)
    gstname = gststruct.get_name()
    source_bin = data
    features = caps.get_features(0)

    # Need to check if the pad created by the decodebin is for video and not
    # audio.
    print("gstname=", gstname)
    if (gstname.find("video") != -1):
        # Link the decodebin pad only if decodebin has picked nvidia
        # decoder plugin nvdec_*. We do this by checking if the pad caps contain
        # NVMM memory features.
        print("features=", features)
        if features.contains("memory:NVMM"):
            # Get the source bin ghost pad
            bin_ghost_pad = source_bin.get_static_pad("src")
            if not bin_ghost_pad.set_target(decoder_src_pad):
                sys.stderr.write("Failed to link decoder src pad to source bin ghost pad\n")
            sys.stderr.write(" Error: Decodebin did not pick nvidia decoder plugin.\n")

def decodebin_child_added(child_proxy, Object, name, user_data):

      print("Decodebin child added:", name)
      if(name.find("decodebin") != -1):
          Object.connect("child-added", decodebin_child_added, user_data)
      if(name.find("nvv4l2decoder") != -1):
          print("Seting bufapi_version\n")
      if use_tcp and "source" in name:
          source_element = child_proxy.get_by_name("source")
          if source_element.find_property('protocols') != None:
              Object.set_property("protocols", GstRtsp.RTSPLowerTrans.TCP)

def create_source_bin(index, uri):
      bin_name = "source-bin-%02d" % index
      nbin =
      if not nbin:
          sys.stderr.write(" Unable to create source bin \n")

      uri_decode_bin = Gst.ElementFactory.make("uridecodebin", "uri-decode-bin")
      if not uri_decode_bin:
          sys.stderr.write(" Unable to create uri decode bin \n")

      uri_decode_bin.connect("pad-added", cb_newpad, nbin)
      uri_decode_bin.connect("child-added", decodebin_child_added, nbin)
      Gst.Bin.add(nbin, uri_decode_bin)
      bin_pad = nbin.add_pad(Gst.GhostPad.new_no_target("src", Gst.PadDirection.SRC))
      if not bin_pad:
          sys.stderr.write(" Failed to add ghost pad in source bin \n")
          return None
      return nbin

Added Videorate filter also

  def create_videorate_filter(cam_id, intended_fps):
      filter = Gst.ElementFactory.make('videorate', f'videorate_{cam_id}')
      filter.set_property("drop-only", True)
      filter.set_property("max-rate", int(intended_fps))
      if not filter:
          sys.stderr.write(" Unable to create capsfilter \n")
      return filter
  1. You can check the latency in your pipeline by referring to the FAQ.
  2. You can use the nvidia-smi dmon command to check the loading of your GPU.

If loading(GPU, Memory, dec…) has reached its limit or the maximum latency of one of the plugins is 10ms, this means that the maximum fps is 100.

We are running multiple pipelines, each pipeline handles 2 cameras

does 10ms latency value still hold?

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks

Yes. 10ms means the latency of any plugin in the pipeline. Have you tried the FAQ and get the latency?

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.