I’m trying to install the TensorRT 6.0.1.8 on my Ubuntu 18.04. I am following https://docs.nvidia.com/deeplearning/sdk/tensorrt-install-guide/index.html#gettingstarted for the installation process and everything went well. Then I tried to verify the installation by running the sampleMNIST. However, the make failed with the following message:
../Makefile.config:7: CUDA_INSTALL_DIR variable is not specified, using /usr/local/cuda by default, use CUDA_INSTALL_DIR=<cuda_directory> to change.
../Makefile.config:10: CUDNN_INSTALL_DIR variable is not specified, using $CUDA_INSTALL_DIR by default, use CUDNN_INSTALL_DIR=<cudnn_directory> to change.
Cleaning...
if [ -d ../../bin/chobj/../common ]; then rm -rf ../../bin/chobj/../common; fi; :
if [ -d ../../bin/dchobj/../common ]; then rm -rf ../../bin/dchobj/../common; fi; :
../Makefile.config:7: CUDA_INSTALL_DIR variable is not specified, using /usr/local/cuda by default, use CUDA_INSTALL_DIR=<cuda_directory> to change.
../Makefile.config:10: CUDNN_INSTALL_DIR variable is not specified, using $CUDA_INSTALL_DIR by default, use CUDNN_INSTALL_DIR=<cudnn_directory> to change.
if [ ! -d ../../bin/chobj/../common ]; then mkdir -p ../../bin/dchobj/../common; fi; :
Compiling: sampleMNIST.cpp
Linking: ../../bin/sample_mnist_debug
/usr/bin/ld: warning: libcublas.so.10, needed by ../../lib/libnvinfer.so, not found (try using -rpath or -rpath-link)
/usr/bin/ld: warning: libcudart.so.10.2, needed by ../../lib/libnvinfer.so, not found (try using -rpath or -rpath-link)
../../lib/libnvinfer.so: undefined reference to `cublasHgemm@libcublas.so.10'
../../lib/libnvinfer.so: undefined reference to `cudaEventDestroy@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cudaMalloc@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cudaGetDevice@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cublasHgemmStridedBatched@libcublas.so.10'
../../lib/libnvinfer.so: undefined reference to `cudaMemcpyToSymbol@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cudaRuntimeGetVersion@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cudaLaunchKernel@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cudaGetDeviceProperties@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cudaStreamAddCallback@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cudaFuncSetAttribute@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cudaMemcpy@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cudaCreateChannelDesc@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cublasSgemm_v2@libcublas.so.10'
../../lib/libnvinfer.so: undefined reference to `cublasDestroy_v2@libcublas.so.10'
../../lib/libnvinfer.so: undefined reference to `cublasSgemmEx@libcublas.so.10'
../../lib/libnvinfer.so: undefined reference to `cudaGetLastError@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cudaCreateTextureObject@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cudaGetErrorString@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cublasCreate_v2@libcublas.so.10'
../../lib/libnvinfer.so: undefined reference to `cudaGetDeviceCount@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `__cudaRegisterFatBinary@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cublasGemmEx@libcublas.so.10'
../../lib/libnvinfer.so: undefined reference to `cudaEventElapsedTime@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cublasGetVersion_v2@libcublas.so.10'
../../lib/libnvinfer.so: undefined reference to `cudaFree@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cudaEventCreateWithFlags@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cublasSgemmStridedBatched@libcublas.so.10'
../../lib/libnvinfer.so: undefined reference to `__cudaRegisterFunction@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cublasSetStream_v2@libcublas.so.10'
../../lib/libnvinfer.so: undefined reference to `cudaEventRecord@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cudaStreamQuery@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cudaStreamCreate@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cudaDestroyTextureObject@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cudaStreamDestroy@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cudaDeviceSynchronize@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cudaMemcpyAsync@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cudaDeviceGetAttribute@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `__cudaPushCallConfiguration@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cublasGetMathMode@libcublas.so.10'
../../lib/libnvinfer.so: undefined reference to `__cudaRegisterFatBinaryEnd@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `__cudaUnregisterFatBinary@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cudaEventSynchronize@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cudaMemsetAsync@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `__cudaPopCallConfiguration@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cublasSetMathMode@libcublas.so.10'
../../lib/libnvinfer.so: undefined reference to `cudaMemGetInfo@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cudaStreamSynchronize@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `__cudaRegisterVar@libcudart.so.10.2'
../../lib/libnvinfer.so: undefined reference to `cudaMemcpy2DAsync@libcudart.so.10.2'
collect2: error: ld returned 1 exit status
../Makefile.config:161: recipe for target '../../bin/sample_mnist_debug' failed
make: *** [../../bin/sample_mnist_debug] Error 1
It seems like I don’t have the CUDA 10.2 installed, but when I ran nvidia-smi, the following message confirms that I have 10.2
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 440.48.02 Driver Version: 440.48.02 CUDA Version: 10.2 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 GeForce RTX 208... Off | 00000000:01:00.0 On | N/A |
| 41% 37C P5 36W / 260W | 3483MiB / 11016MiB | 1% Default |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| 0 3310 G /usr/lib/xorg/Xorg 40MiB |
| 0 3343 G /usr/bin/gnome-shell 123MiB |
| 0 4396 G gnome-control-center 38MiB |
| 0 4959 G /usr/lib/xorg/Xorg 1231MiB |
| 0 5076 G /usr/bin/gnome-shell 484MiB |
| 0 5491 G ...quest-channel-token=1054966862813480335 153MiB |
| 0 6019 G ...quest-channel-token=3470144649993740052 938MiB |
| 0 23769 G ...uest-channel-token=10751755228895556740 467MiB |
+-----------------------------------------------------------------------------+
I downloaded TensorRT-6.0.1.8.Ubuntu-18.04.x86_64-gnu.cuda-10.2.cudnn7.6.tar.gz, the cudnn is at 7.6 as I have verified.