Need clarification about TensorRT releases

AI & Data Science Deep Learning (Training & Inference) TensorRT
1

Description

Hi
I have few questions related tensorrt versions…

  1. Why i am seeing only four versions of tensorrt when i run the beow commands…

commands:-

sudo yum-config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/rhel7/x86_64/cuda-rhel7.repo
sudo yum --showduplicate list tensorrt

output;-

tensorrt.x86_64 8.4.1.5-1.cuda10.2 cuda-rhel7-x86_64
tensorrt.x86_64 8.4.1.5-1.cuda11.6 cuda-rhel7-x86_64
tensorrt.x86_64 8.4.2.4-1.cuda10.2 cuda-rhel7-x86_64
tensorrt.x86_64 8.4.2.4-1.cuda11.6 cuda-rhel7-x86_64

  1. How do I get access to older tensorrt packages? so that i can install it with the below command.
    sudo yum install tensorrt-8.4.1.5-1.cuda11.6.x86_64 -y

  2. Latest NVIDIA TensorRT container release 22.7 is with TensorRT 8.4.1.
    Let us say if I build tensorrt engine on 8.4.1.5 (with the rpm installation shown above).
    Can I run inference with the tensorrt engine (version:8.4.1.5) on tensorrt container (release:22.7, tensorrt verison:8.4.1)
    is the difference between 8.4.1.5 and 8.4.1 is ok? or it is not allowed?

Environment

TensorRT Version: n/a
GPU Type: n/a
Nvidia Driver Version: n/a
CUDA Version: n/a
CUDNN Version: n/a
Operating System + Version:n/a
Python Version (if applicable): n/a
TensorFlow Version (if applicable):n/a
PyTorch Version (if applicable): n/a
Baremetal or Container (if container which image + tag):n/a

Relevant Files

n/a

Steps To Reproduce

n/a

2

Hi,
Please refer to the installation steps from the below link if in case you are missing on anything

Also, we suggest you to use TRT NGC containers to avoid any system dependency related issues.

Thanks!

4

Hi,

You can download the older packages from the below link and follow the instructions from the installation guide above shared.

TensorRT engine is specific to the TensorRT version, GPU, OS, platform, etc, and not portable across the versions/platforms. You need to rebuild the engine to run on different version.

Thank you.

5

Hi spolisetty

  1. If I see NVIDIA NGC catalog(below link) for tensorrt images, there is no separate tags for different OS’s.
    TensorRT | NVIDIA NGC
    So I am thinking that tensorrt is not OS dependent (I meant different types of linux: Debian, centos…)
    Can you please confirm on this…
    This is important to know for me because I can only build my tensorrt engine on CentOs7 and I have to run inference using Nvidia Tensorrt container…This is infrastructure limitation that I have.

  2. If I see the release notes of tensorrt in below link: version 8.4.1 is present, but version 8.4.1.5 is not.
    Release Notes :: NVIDIA Deep Learning TensorRT Documentation
    And when I try to download tensorrt from the below link, by selecting “TensorRT 8.4 GA for X86_64 architecture”, for Centos 7 and cuda 11.XX, it is downloading version 8.4.1.5.
    And there is no link that just gives “TensorRT 8.4.1”, without the fourth digit.
    What version of tensorRT is actually in tensorrt container (release:22.7, tensorrt verison:8.4.1)? I meant the fourth digit of tensorrt version?

Thanks
Ramachandra

6

Hi,

  1. Please refer to the NGC container release notes for more details, as mentioned previously, TensorRT is platform-dependent, you need to re-build the engine on the platform you’d like to run. Container Release Notes :: NVIDIA Deep Learning TensorRT Documentation
  2. They both are the same version 8.4.1.5. You can try dpkg -l | grep TensorRT inside the container for confirmation.

Thank you.

7

Thanks…that is useful…

8

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