ubuntu 20.04 + cuda-11.8 + cudnn-8.6+TensorRT-8.6

1、装显卡驱动

ubuntu20.04 + cuda10.0 + cudnn7.6.4_我是谁？？的博客-CSDN博客

查看支持的驱动版本：

查看本机显卡能够配置的驱动信息

lu@host:/usr/local$ ubuntu-drivers devices
== /sys/devices/pci0000:00/0000:00:01.0/0000:01:00.0 ==
modalias : pci:v000010DEd000021C4sv000010DEsd000021C4bc03sc00i00
vendor   : NVIDIA Corporation
model    : TU116 [GeForce GTX 1660 SUPER]
driver   : nvidia-driver-525-open - distro non-free
driver   : nvidia-driver-450-server - distro non-free
driver   : nvidia-driver-525 - distro non-free
driver   : nvidia-driver-525-server - distro non-free
driver   : nvidia-driver-535-open - distro non-free
driver   : nvidia-driver-470-server - distro non-free
driver   : nvidia-driver-535-server-open - distro non-free recommended
driver   : nvidia-driver-535-server - distro non-free
driver   : nvidia-driver-535 - distro non-free
driver   : nvidia-driver-470 - distro non-free
driver   : xserver-xorg-video-nouveau - distro free builtin

lu@host:/usr/local$ 

安装驱动：

我这边安装的是nvidia-driver-535

sudo apt install nvidia-driver-535

重启电脑，查看nvidia驱动：

lu@host:/usr/local$ nvidia-smi 
Fri Nov  3 00:26:46 2023       
+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 535.113.01             Driver Version: 535.113.01   CUDA Version: 12.2     |
|-----------------------------------------+----------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |         Memory-Usage | GPU-Util  Compute M. |
|                                         |                      |               MIG M. |
|=========================================+======================+======================|
|   0  NVIDIA GeForce GTX 1660 ...    Off | 00000000:01:00.0 Off |                  N/A |
| 29%   34C    P8              N/A /  N/A |    348MiB /  6144MiB |     33%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+
                                                                                         
+---------------------------------------------------------------------------------------+
| Processes:                                                                            |
|  GPU   GI   CI        PID   Type   Process name                            GPU Memory |
|        ID   ID                                                             Usage      |
|=======================================================================================|
|    0   N/A  N/A       890      G   /usr/lib/xorg/Xorg                           45MiB |
|    0   N/A  N/A      1432      G   /usr/lib/xorg/Xorg                          129MiB |
|    0   N/A  N/A      1667      G   /usr/bin/gnome-shell                         30MiB |
|    0   N/A  N/A      2352      G   ...3584735,16244303988823860755,262144      131MiB |
+---------------------------------------------------------------------------------------+
lu@host:/usr/local$ 

可以看到535最高支持cuda-12.2版本,我这里安装cuda-11.8显然满足要求(驱动版本可以高于对应cuda版本)。

2、安装CUDA

下载cuda:

链接：CUDA Toolkit Archive | NVIDIA Developer

CUDA推荐下载.run可以根据提示安装,执行如下命令：

sudo sh cuda_11.8.0_520.61.05_linux.run

出现选择：选择Continue

光标分别移动到Driver、CUDA Demo Suite 11.8、CUDA Documentation 11.8，按空格键，去掉安装选择，如下图，只安装CUDA Toolkit 11.8

设置cuda环境变量：

打开主目录下的 .bashrc文件添加如下路径，例如我的.bashrc文件在/home/lu/下。
 
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda-11.8/lib64
export PATH=$PATH:/usr/local/cuda-11.8/bin
export CUDA_HOME=$CUDA_HOME:/usr/local/cuda-11.8

终端运行：source ~/.bashrc
 
检查：nvcc --version

3.cudnn的安装

下载安装文件：

按需求下载cudnn的安装文件：cuDNN Archive | NVIDIA Developer

安装cudnn：

这里我下载的cudnn-linux-x86_64-8.6.0.163_cuda11-archive.tar.xz

解压下载的文件，可以看到cuda文件夹，在当前目录打开终端，执行如下命令：

   sudo cp cuda/include/cudnn* /usr/local/cuda/include/
     
   sudo cp cuda/lib64/libcudnn* /usr/local/cuda/lib64/
     
   sudo chmod a+r /usr/local/cuda/include/cudnn*
     
   sudo chmod a+r /usr/local/cuda/lib64/libcudnn*

验证是否安装成功：

sudo cat  /home/cuda/cuda118/include/cudnn_version.h | grep CUDNN_MAJOR -A 2

4.TensorRT安装

下载：

网址：Log in | NVIDIA Developer

选择TensorTR8.6 GA,我下载的版本是TensorRT-8.6.1.6.Linux.x86_64-gnu.cuda-11.8.tar.gz

安装：

1.解压 tar -xzvf TensorRT-8.6.1.6.Linux.x86_64-gnu.cuda-11.8.tar.gz 

2.执行命令sudo cp -rf TensorRT-8.6.1.6 / 将解压的内容拷贝到指定目录（我是到/usr/local/目录）

配置TensorRT:

打开主目录下的 .bashrc文件添加如下路径，例如我的.bashrc文件在/home/lu/下。
 
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/TensorRT-8.6.1.6/lib
export C_INCLUDE_PATH=$C_INCLUDE_PATH:/usr/local/TensorRT-8.6.1.6/include
export CPLUS_INCLUDE_PATH=$CPLUS_INCLUDE_PATH:/usr/local/TensorRT-8.6.1.6/include

终端运行：source ~/.bashrc

注意：也可以参考cudnn的安装方式，直接把头文件和库拷贝到cuda目录下，这样就不要再配置环境变量了。

测试TensorRT安装是否成功

# 进入到/usr/local/TensorRT-8.6.1.6/samples/sampleOnnxMNIST目录
cd /usr/local/TensorRT-8.6.1.6/samples/sampleOnnxMNIST

# 执行make命令编译
make

# 会在/usr/local/TensorRT-8.6.1.6/bin/目录下生成sample_onnx_mnist文件
/usr/local/TensorRT-8.6.1.6/bin/sample_onnx_mnist

成功后显示如下：

lu@host:/usr/local/TensorRT-8.6.1.6/samples/sampleOnnxMNIST$ /usr/local/TensorRT-8.6.1.6/bin/sample_onnx_mnist
&&&& RUNNING TensorRT.sample_onnx_mnist [TensorRT v8601] # /usr/local/TensorRT-8.6.1.6/bin/sample_onnx_mnist
[11/03/2023-00:57:21] [I] Building and running a GPU inference engine for Onnx MNIST
[11/03/2023-00:57:22] [I] [TRT] [MemUsageChange] Init CUDA: CPU +14, GPU +0, now: CPU 19, GPU 448 (MiB)
[11/03/2023-00:57:28] [I] [TRT] [MemUsageChange] Init builder kernel library: CPU +897, GPU +174, now: CPU 992, GPU 586 (MiB)
[11/03/2023-00:57:28] [I] [TRT] ----------------------------------------------------------------
[11/03/2023-00:57:28] [I] [TRT] Input filename:   ../../data/mnist/mnist.onnx
[11/03/2023-00:57:28] [I] [TRT] ONNX IR version:  0.0.3
[11/03/2023-00:57:28] [I] [TRT] Opset version:    8
[11/03/2023-00:57:28] [I] [TRT] Producer name:    CNTK
[11/03/2023-00:57:28] [I] [TRT] Producer version: 2.5.1
[11/03/2023-00:57:28] [I] [TRT] Domain:           ai.cntk
[11/03/2023-00:57:28] [I] [TRT] Model version:    1
[11/03/2023-00:57:28] [I] [TRT] Doc string:       
[11/03/2023-00:57:28] [I] [TRT] ----------------------------------------------------------------
[11/03/2023-00:57:28] [W] [TRT] onnx2trt_utils.cpp:374: Your ONNX model has been generated with INT64 weights, while TensorRT does not natively support INT64. Attempting to cast down to INT32.
[11/03/2023-00:57:28] [I] [TRT] BuilderFlag::kTF32 is set but hardware does not support TF32. Disabling TF32.
[11/03/2023-00:57:28] [I] [TRT] Graph optimization time: 0.0039157 seconds.
[11/03/2023-00:57:28] [I] [TRT] BuilderFlag::kTF32 is set but hardware does not support TF32. Disabling TF32.
[11/03/2023-00:57:28] [I] [TRT] Local timing cache in use. Profiling results in this builder pass will not be stored.
[11/03/2023-00:57:29] [I] [TRT] Detected 1 inputs and 1 output network tensors.
[11/03/2023-00:57:29] [I] [TRT] Total Host Persistent Memory: 24224
[11/03/2023-00:57:29] [I] [TRT] Total Device Persistent Memory: 0
[11/03/2023-00:57:29] [I] [TRT] Total Scratch Memory: 0
[11/03/2023-00:57:29] [I] [TRT] [MemUsageStats] Peak memory usage of TRT CPU/GPU memory allocators: CPU 0 MiB, GPU 4 MiB
[11/03/2023-00:57:29] [I] [TRT] [BlockAssignment] Started assigning block shifts. This will take 6 steps to complete.
[11/03/2023-00:57:29] [I] [TRT] [BlockAssignment] Algorithm ShiftNTopDown took 0.017708ms to assign 3 blocks to 6 nodes requiring 32256 bytes.
[11/03/2023-00:57:29] [I] [TRT] Total Activation Memory: 31744
[11/03/2023-00:57:29] [I] [TRT] [MemUsageChange] TensorRT-managed allocation in building engine: CPU +0, GPU +4, now: CPU 0, GPU 4 (MiB)
[11/03/2023-00:57:29] [I] [TRT] Loaded engine size: 0 MiB
[11/03/2023-00:57:29] [I] [TRT] [MemUsageChange] TensorRT-managed allocation in engine deserialization: CPU +0, GPU +0, now: CPU 0, GPU 0 (MiB)
[11/03/2023-00:57:30] [I] [TRT] [MemUsageChange] TensorRT-managed allocation in IExecutionContext creation: CPU +0, GPU +0, now: CPU 0, GPU 0 (MiB)
[11/03/2023-00:57:30] [I] Input:
[11/03/2023-00:57:30] [I] @@@@@@@@@@@@@@@@@@@@@@@@@@@@
@@@@@@@@@@@@@@@@@@@@@@@@@@@@
@@@@@@@@@@@@@%.:@@@@@@@@@@@@
@@@@@@@@@@@@@: *@@@@@@@@@@@@
@@@@@@@@@@@@* =@@@@@@@@@@@@@
@@@@@@@@@@@% :@@@@@@@@@@@@@@
@@@@@@@@@@@- *@@@@@@@@@@@@@@
@@@@@@@@@@# .@@@@@@@@@@@@@@@
@@@@@@@@@@: #@@@@@@@@@@@@@@@
@@@@@@@@@+ -@@@@@@@@@@@@@@@@
@@@@@@@@@: %@@@@@@@@@@@@@@@@
@@@@@@@@+ +@@@@@@@@@@@@@@@@@
@@@@@@@@:.%@@@@@@@@@@@@@@@@@
@@@@@@@% -@@@@@@@@@@@@@@@@@@
@@@@@@@% -@@@@@@#..:@@@@@@@@
@@@@@@@% +@@@@@-    :@@@@@@@
@@@@@@@% =@@@@%.#@@- +@@@@@@
@@@@@@@@..%@@@*+@@@@ :@@@@@@
@@@@@@@@= -%@@@@@@@@ :@@@@@@
@@@@@@@@@- .*@@@@@@+ +@@@@@@
@@@@@@@@@@+  .:-+-: .@@@@@@@
@@@@@@@@@@@@+:    :*@@@@@@@@
@@@@@@@@@@@@@@@@@@@@@@@@@@@@
@@@@@@@@@@@@@@@@@@@@@@@@@@@@
@@@@@@@@@@@@@@@@@@@@@@@@@@@@
@@@@@@@@@@@@@@@@@@@@@@@@@@@@
@@@@@@@@@@@@@@@@@@@@@@@@@@@@
@@@@@@@@@@@@@@@@@@@@@@@@@@@@

[11/03/2023-00:57:30] [I] Output:
[11/03/2023-00:57:30] [I]  Prob 0  0.0000 Class 0: 
[11/03/2023-00:57:30] [I]  Prob 1  0.0000 Class 1: 
[11/03/2023-00:57:30] [I]  Prob 2  0.0000 Class 2: 
[11/03/2023-00:57:30] [I]  Prob 3  0.0000 Class 3: 
[11/03/2023-00:57:30] [I]  Prob 4  0.0000 Class 4: 
[11/03/2023-00:57:30] [I]  Prob 5  0.0000 Class 5: 
[11/03/2023-00:57:30] [I]  Prob 6  1.0000 Class 6: **********
[11/03/2023-00:57:30] [I]  Prob 7  0.0000 Class 7: 
[11/03/2023-00:57:30] [I]  Prob 8  0.0000 Class 8: 
[11/03/2023-00:57:30] [I]  Prob 9  0.0000 Class 9: 
[11/03/2023-00:57:30] [I] 
&&&& PASSED TensorRT.sample_onnx_mnist [TensorRT v8601] # /usr/local/TensorRT-8.6.1.6/bin/sample_onnx_mnist
lu@host:/usr/local/TensorRT-8.6.1.6/samples/sampleOnnxMNIST$