Skip to content

Commit a860e49

Browse files
copybara-githubcopybara-github
committed
Merge pull request #2055 from nluehr:cuda-repo-update
PiperOrigin-RevId: 446640821
2 parents 20f4caa + 7c660f6 commit a860e49

File tree

1 file changed

+27
-48
lines changed

1 file changed

+27
-48
lines changed

‎site/en/install/gpu.md

Lines changed: 27 additions & 48 deletions
Original file line numberDiff line numberDiff line change
@@ -67,7 +67,7 @@ The following NVIDIA® software must be installed on your system:
6767
* [cuDNN SDK 8.1.0](https://developer.nvidia.com/cudnn){:.external}
6868
[cuDNN versions](https://developer.nvidia.com/rdp/cudnn-archive){:.external}).
6969
* *(Optional)*
70-
[TensorRT 7](https://docs.nvidia.com/deeplearning/tensorrt/archives/index.html#trt_7){:.external}
70+
[TensorRT](https://docs.nvidia.com/deeplearning/tensorrt/archives/index.html#trt_7){:.external}
7171
to improve latency and throughput for inference on some models.
7272

7373
## Linux setup
@@ -93,75 +93,54 @@ This section shows how to install CUDA® 11 (TensorFlow >= 2.4.0) on Ubuntu
9393
Caution: [Secure Boot](https://wiki.ubuntu.com/UEFI/SecureBoot){:.external}
9494
complicates installation of the NVIDIA driver and is beyond the scope of these instructions.
9595

96-
97-
#### Ubuntu 18.04 (CUDA 11.0)
96+
#### Ubuntu 18.04 (CUDA 11.2)
9897

9998
<pre class="prettyprint lang-bsh">
10099
# Add NVIDIA package repositories
101-
<code class="devsite-terminal">wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/cuda-ubuntu1804.pin</code>
102-
<code class="devsite-terminal">sudo mv cuda-ubuntu1804.pin /etc/apt/preferences.d/cuda-repository-pin-600</code>
103-
<code class="devsite-terminal">sudo apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/7fa2af80.pub</code>
104-
<code class="devsite-terminal">sudo add-apt-repository "deb https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/ /"</code>
105-
<code class="devsite-terminal">sudo apt-get update</code>
106-
107-
<code class="devsite-terminal">wget http://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1804/x86_64/nvidia-machine-learning-repo-ubuntu1804_1.0.0-1_amd64.deb</code>
108-
109-
<code class="devsite-terminal">sudo apt install ./nvidia-machine-learning-repo-ubuntu1804_1.0.0-1_amd64.deb</code>
110-
<code class="devsite-terminal">sudo apt-get update</code>
111-
112-
<code class="devsite-terminal">wget https://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1804/x86_64/libnvinfer7_7.1.3-1+cuda11.0_amd64.deb</code>
113-
<code class="devsite-terminal">sudo apt install ./libnvinfer7_7.1.3-1+cuda11.0_amd64.deb</code>
100+
# Note: For the Ubuntu version other than 18.04 or CPU architecture other than x86,
101+
# replace `ubuntu1804` and/or `x86_64` as needed in the following URL.
102+
<code class="devsite-terminal">wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/cuda-keyring_1.0-1_all.deb</code>
103+
<code class="devsite-terminal">sudo dpkg -i cuda-keyring_1.0-1_all.deb</code>
114104
<code class="devsite-terminal">sudo apt-get update</code>
115105

116106
# Install development and runtime libraries (~4GB)
107+
# libnvinfer packages are optional, needed to support TensorRT inference.
117108
<code class="devsite-terminal">sudo apt-get install --no-install-recommends \
118-
cuda-11-0 \
119-
libcudnn8=8.0.4.30-1+cuda11.0 \
120-
libcudnn8-dev=8.0.4.30-1+cuda11.0
109+
cuda-11-2 \
110+
libcudnn8=8.1.0.77-1+cuda11.2 \
111+
libcudnn8-dev=8.1.0.77-1+cuda11.2 \
112+
libnvinfer8=8.2.4-1+cuda11.4 \
113+
libnvinfer-dev=8.2.4-1+cuda11.4 \
114+
libnvinfer-plugin8=8.2.4-1+cuda11.4 \
115+
libnvinfer-plugin-dev=8.2.4-1+cuda11.4
121116
</code>
122-
# Reboot. Check that GPUs are visible using the command: nvidia-smi
123117

124-
# Install TensorRT. Requires that libcudnn8 is installed above.
125-
<code class="devsite-terminal">sudo apt-get install -y --no-install-recommends libnvinfer7=7.1.3-1+cuda11.0 \
126-
libnvinfer-dev=7.1.3-1+cuda11.0 \
127-
libnvinfer-plugin7=7.1.3-1+cuda11.0
128-
</code>
118+
# Reboot. Check that GPUs are visible using the command: nvidia-smi
129119
</pre>
130120

131-
#### Ubuntu 16.04 (CUDA 11.0)
121+
#### Ubuntu 16.04 (CUDA 11.2)
132122

133123
<pre class="prettyprint lang-bsh">
134124
# Add NVIDIA package repositories
135-
# Add HTTPS support for apt-key
136-
<code class="devsite-terminal">sudo apt-get install gnupg-curl</code>
137-
<code class="devsite-terminal">wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64/cuda-ubuntu1604.pin</code>
138-
<code class="devsite-terminal">sudo mv cuda-ubuntu1604.pin /etc/apt/preferences.d/cuda-repository-pin-600</code>
139-
<code class="devsite-terminal">sudo apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64/7fa2af80.pub</code>
140-
<code class="devsite-terminal">sudo add-apt-repository "deb https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64/ /"</code>
141125
<code class="devsite-terminal">sudo apt-get update</code>
142-
<code class="devsite-terminal">wget https://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1604/x86_64/nvidia-machine-learning-repo-ubuntu1604_1.0.0-1_amd64.deb</code>
143-
<code class="devsite-terminal">sudo apt install ./nvidia-machine-learning-repo-ubuntu1604_1.0.0-1_amd64.deb</code>
144-
<code class="devsite-terminal">sudo apt-get update</code>
145-
<code class="devsite-terminal">wget https://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1604/x86_64/libnvinfer7_7.1.3-1+cuda11.0_amd64.deb</code>
146-
<code class="devsite-terminal">sudo apt install ./libnvinfer7_7.1.3-1+cuda11.0_amd64.deb</code>
126+
<code class="devsite-terminal">sudo apt-get install -y apt-transport-https</code>
127+
<code class="devsite-terminal">wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64/cuda-keyring_1.0-1_all.deb</code>
128+
<code class="devsite-terminal">sudo dpkg -i cuda-keyring_1.0-1_all.deb</code>
147129
<code class="devsite-terminal">sudo apt-get update</code>
148130

149131
# Install development and runtime libraries (~4GB)
132+
# libnvinfer packages are optional, needed to support TensorRT inference.
150133
<code class="devsite-terminal">sudo apt-get install --no-install-recommends \
151-
cuda-11-0 \
152-
libcudnn8=8.0.4.30-1+cuda11.0 \
153-
libcudnn8-dev=8.0.4.30-1+cuda11.0
134+
cuda-11-2 \
135+
libcudnn8=8.1.0.77-1+cuda11.2 \
136+
libcudnn8-dev=8.1.0.77-1+cuda11.2 \
137+
libnvinfer8=8.0.1-1+cuda11.3 \
138+
libnvinfer-dev=8.0.1-1+cuda11.3 \
139+
libnvinfer-plugin8=8.0.1-1+cuda11.3 \
140+
libnvinfer-plugin-dev=8.0.1-1+cuda11.3
154141
</code>
155142

156143
# Reboot. Check that GPUs are visible using the command: nvidia-smi
157-
158-
# Install TensorRT. Requires that libcudnn7 is installed above.
159-
<code class="devsite-terminal">sudo apt-get install -y --no-install-recommends \
160-
libnvinfer7=7.1.3-1+cuda11.0 \
161-
libnvinfer-dev=7.1.3-1+cuda11.0 \
162-
libnvinfer-plugin7=7.1.3-1+cuda11.0 \
163-
libnvinfer-plugin-dev=7.1.3-1+cuda11.0
164-
</code>
165144
</pre>
166145

167146

0 commit comments

Comments
 (0)