Can anyone point me to appropriate tested steps or at least a comprehensive flow and compatible versions for installing a Tesla P4 through the DeepStream SDK on Ubuntu 18.04? I assume this would include the CUDA toolkit and TensorRT as well.
I started with the nvidia-418 tesla p4 driver and thought I was following readme and docs appropriately to install the other pieces, but when I got to trying to follow the Deepstream samples, missing libraries and mismatched versions of such began showing up. If I have to tear down to a base image and start over, that’s fine, but I’d like to know I’m not making the same mistakes (if there were such) in the process.
Yes. It needs cuda, tensorRT firstly.
Can you check the README, this is the deepstream 4.0 pre-requistites in README:
Setup pre-requisites:
- Ubuntu 18.04
- Gstreamer 1.14.1
- NVIDIA driver 418+
- CUDA 10.1
- TensorRT 5.1.2 RC
Yes the README for DeepStream 3.0 pre-reqs according the web page are:
Verify that you have the following installed on your system
Ubuntu 16.04 LTS (with GCC 5.4)
NVIDIA Display Driver ver. 410.72
NVIDIA Video SDK 8.2
NVIDIA cuDNN 7.3
NVIDIA TensorRT 5.0
NVIDIA CUDA 10.0
Download the DeepStream 3.0 tar file (below)
Extract the package at the location of your choice
Follow the usage instructions from the Deepstream_User_guide.pdf in the downloaded package.
However I’m starting on Ubuntu 18.04, with GCC 7.4. So do I load the 410.72 driver or the 418 driver? I note that the DeepStream 3.0 reqs don’t even mention a version of Gstreamer (I have 1.0 installed). Since my first time I tried the 418 driver and CUDA 10.1 (basically all the current versions except I didn’t find DeepStream 4.0), that I should probably back down to the listed versions even though I’m on Ubuntu 18 and updated GCC?
Another confusion is the discrepancy between the information at https://developer.nvidia.com/compute/machine-learning/deepstream-downloads (in previous post) with the README in the 3.0 download:
Setup pre-requisites:
- Ubuntu 16.04
- Gstreamer 1.8.3
- NVIDIA driver 410+
- CUDA 10.0
- TensorRT 5.0
The listed versions are the same but the README doesn’t mention the Video SDK and cuDNN.
Also the file “Deepstream_User_guide.pdf” referenced on the web page does not exist in either the SDK download or the documentation download.
Please follow REAME.
3.0 does not support Ubuntu 18.04
Don’t need video codec sdk. libnvcuvid.so will be there after nvidia driver installation.
tensorRT version will specify the version of cudnn.
Since 4.0 is not yet released/available (that I could find), it would appear my only supported option is to back down to Ubuntu 16.04. That is disappointing. Thank you.
Hi 4.0 will be released soon, maybe in 3 weeks.
Hey Chris, any idea when they will release v4? So many of us are stuck between both worlds where we can’t downgrade to Ubuntu 16 and can’t use Deepstream 3…
Sorry. It should be next week.