Use gpu-id=1 in deepstream-app

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

Please provide complete information as applicable to your setup.

• Hardware Platform (Jetson / GPU) GPU Tesla T4
• DeepStream Version Deepstream SDK 5.0

Hi,

I am trying to use the second GPU on a server for running an instance of deepstream-app pipeline. I am using the attached config files, however I keep getting the following error:

ERROR from src_bin_muxer: Memory Compatibility Error:Input surface gpu-id doesnt match with configured gpu-id for element, please allocate input using unified memory, or use same gpu-ids OR, if same gpu-ids are used ensure appropriate Cuda memories are used
Debug info: gstnvstreammux.c(1224): copy_data_cuda (): /GstPipeline:pipeline/GstBin:multi_src_bin/GstNvStreamMux:src_bin_muxer:
surface-gpu-id=0,src_bin_muxer-gpu-id=1
0:00:04.824427829  7547 0x5596b455d2d0 WARN                 nvinfer gstnvinfer.cpp:1240:convert_batch_and_push_to_input_thread:<primary_gie> error: NvBufSurfTransform failed with error -1 while converting buffer
ERROR from primary_gie: NvBufSurfTransform failed with error -1 while converting buffer
Debug info: gstnvinfer.cpp(1240): convert_batch_and_push_to_input_thread (): /GstPipeline:pipeline/GstBin:primary_gie_bin/GstNvInfer:primary_gie
Quitting
App run failed

I have tried changing cudadec-memtype=2 and nvbuf-memory-type=3 under [streammux] with no luck. I have tried setting nvbuf-memory-type=3 under all plugins as well but didn’t work. Based on the error it seems like surface-gpu-id=0 no matter what I set in the configuration file, so how do I change this value?

Would you please let me know what is the problem? (I have to ensure that the machine indeed has
config_infer_primary_yoloV4.txt (2.9 KB)
d_gpu1.txt (3.8 KB)
2 GPUs).

3

I tried the same code on another machine with two GPU and it works fine. Do you know what can cause this problem? it seems like there is a specific problem with the machine, not the deepstream configuration.

4

Quick update, I realized that I am getting this error only if the source file is a small video file (a few seconds) but I don’t get this error when it is a larger video file, and it happens on both servers. Any suggestions on how I can fix this?

5

Hi @MGh

use the second GPU on a server for running an instance of deepstream-app pipeline
After you deepstream-app pipeline works on GPU#0, you can use “export CUDA_VISIABLE_DEVICES=1” or “CUDA_VISIABLE_DEVICES=1 deepstream-app …” to run your application on the 2nd GPU, by this, you don’t need to modify anything in your app or config

6

Hi @mchi

Thanks a lot for your response. I tried what you suggested however even though I set CUDA_VISIABLE_DEVICES=1 when I check with nvidia-smi still GPU 0 is being used to capacity and GPU 1 is idle.

7

There are so many GPU users who can use “CUDA_VISIBLE_DEVICES” without any issue, I believe it can work well.
You can search on Ethernet about the usage to find some clues about why it failed on your side.

8

I see, sure I will do that. Just one question, does that make sense that it doesn’t work for short video files but it does work for larger input video files?

9

this is not expected

10

Sorry!
Note, it’s CUDA_VISIBLE_DEVICES not CUDA_VISIABLE_DEVICES .

11

@mchi Thank you, this works!

1 Like
13

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