You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: docs/_latest/deeplearning4j-config-cudnn.md
+23-58Lines changed: 23 additions & 58 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -10,12 +10,12 @@ weight: 3
10
10
11
11
Deeplearning4j supports CUDA but can be further accelerated with cuDNN. Most 2D CNN layers (such as ConvolutionLayer, SubsamplingLayer, etc), and also LSTM and BatchNormalization layers support CuDNN.
12
12
13
-
The only thing we need to do to have DL4J load cuDNN is to add a dependency on `deeplearning4j-cuda-9.0`, `deeplearning4j-cuda-9.2`, or `deeplearning4j-cuda-10.0`, for example:
13
+
The only thing we need to do to have DL4J load cuDNN is to add a dependency on `deeplearning4j-cuda-9.2`, `deeplearning4j-cuda-10.0`, `deeplearning4j-cuda-10.1`, or `deeplearning4j-cuda-10.2` for example:
14
14
15
15
```xml
16
16
<dependency>
17
17
<groupId>org.deeplearning4j</groupId>
18
-
<artifactId>deeplearning4j-cuda-9.0</artifactId>
18
+
<artifactId>deeplearning4j-cuda-9.2</artifactId>
19
19
<version>{{page.version}}</version>
20
20
</dependency>
21
21
```
@@ -24,7 +24,7 @@ or
24
24
```xml
25
25
<dependency>
26
26
<groupId>org.deeplearning4j</groupId>
27
-
<artifactId>deeplearning4j-cuda-9.2</artifactId>
27
+
<artifactId>deeplearning4j-cuda-10.0</artifactId>
28
28
<version>{{page.version}}</version>
29
29
</dependency>
30
30
```
@@ -33,11 +33,21 @@ or
33
33
```xml
34
34
<dependency>
35
35
<groupId>org.deeplearning4j</groupId>
36
-
<artifactId>deeplearning4j-cuda-10.0</artifactId>
36
+
<artifactId>deeplearning4j-cuda-10.1</artifactId>
37
37
<version>{{page.version}}</version>
38
38
</dependency>
39
39
```
40
40
41
+
or
42
+
```xml
43
+
<dependency>
44
+
<groupId>org.deeplearning4j</groupId>
45
+
<artifactId>deeplearning4j-cuda-10.2</artifactId>
46
+
<version>{{page.version}}</version>
47
+
</dependency>
48
+
```
49
+
50
+
41
51
The actual library for cuDNN is not bundled, so be sure to download and install the appropriate package for your platform from NVIDIA:
@@ -48,66 +58,21 @@ Note there are multiple combinations of cuDNN and CUDA supported. At this time t
48
58
<th>CUDA Version</th>
49
59
<th>cuDNN Version</th>
50
60
</tr>
51
-
<tr><td>9.0</td><td>7.0</td></tr>
52
-
<tr><td>9.2</td><td>7.1</td></tr>
53
-
<tr><td>10.0</td><td>7.3</td></tr>
61
+
<tr><td>9.2</td><td>7.2</td></tr>
62
+
<tr><td>10.0</td><td>7.4</td></tr>
63
+
<tr><td>10.1</td><td>7.6</td></tr>
64
+
<tr><td>10.2</td><td>7.6</td></tr>
54
65
</table>
55
66
56
67
57
-
To install, simply extract the library to a directory found in the system path used by native libraries. The easiest way is to place it alongside other libraries from CUDA in the default directory (`/usr/local/cuda/lib64/` on Linux, `/usr/local/cuda/lib/` on Mac OS X, and `C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0\bin\`, `C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.2\bin\`, or `C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0\bin\` on Windows).
68
+
To install, simply extract the library to a directory found in the system path used by native libraries. The easiest way is to place it alongside other libraries from CUDA in the default directory (`/usr/local/cuda/lib64/` on Linux, `/usr/local/cuda/lib/` on Mac OS X, and `C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0\bin\`, `C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.1\bin\`, or `C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.2\bin\` on Windows).
58
69
59
-
Alternatively, in the case of CUDA 9.2 or 10.0, cuDNN comes bundled with the "redist" package of the [JavaCPP Presets for CUDA](https://github.com/bytedeco/javacpp-presets/tree/master/cuda). [After agreeing to the license](https://github.com/bytedeco/javacpp-presets/tree/master/cuda#license-agreements), we can add the following dependencies instead of installing CUDA and cuDNN:
70
+
Alternatively, in the case of CUDA 10.2, cuDNN comes bundled with the "redist" package of the [JavaCPP Presets for CUDA](https://github.com/bytedeco/javacpp-presets/tree/master/cuda). [After agreeing to the license](https://github.com/bytedeco/javacpp-presets/tree/master/cuda#license-agreements), we can add the following dependencies instead of installing CUDA and cuDNN:
60
71
61
72
<dependency>
62
-
<groupId>org.bytedeco.javacpp-presets</groupId>
63
-
<artifactId>cuda</artifactId>
64
-
<version>9.2-7.1-1.4.2</version>
65
-
<classifier>linux-x86_64-redist</classifier>
66
-
</dependency>
67
-
<dependency>
68
-
<groupId>org.bytedeco.javacpp-presets</groupId>
69
-
<artifactId>cuda</artifactId>
70
-
<version>9.2-7.1-1.4.2</version>
71
-
<classifier>linux-ppc64le-redist</classifier>
72
-
</dependency>
73
-
<dependency>
74
-
<groupId>org.bytedeco.javacpp-presets</groupId>
75
-
<artifactId>cuda</artifactId>
76
-
<version>9.2-7.1-1.4.2</version>
77
-
<classifier>macosx-x86_64-redist</classifier>
78
-
</dependency>
79
-
<dependency>
80
-
<groupId>org.bytedeco.javacpp-presets</groupId>
81
-
<artifactId>cuda</artifactId>
82
-
<version>9.2-7.1-1.4.2</version>
83
-
<classifier>windows-x86_64-redist</classifier>
84
-
</dependency>
85
-
86
-
or
87
-
88
-
<dependency>
89
-
<groupId>org.bytedeco.javacpp-presets</groupId>
90
-
<artifactId>cuda</artifactId>
91
-
<version>10.0-7.3-1.4.3</version>
92
-
<classifier>linux-x86_64-redist</classifier>
93
-
</dependency>
94
-
<dependency>
95
-
<groupId>org.bytedeco.javacpp-presets</groupId>
96
-
<artifactId>cuda</artifactId>
97
-
<version>10.0-7.3-1.4.3</version>
98
-
<classifier>linux-ppc64le-redist</classifier>
99
-
</dependency>
100
-
<dependency>
101
-
<groupId>org.bytedeco.javacpp-presets</groupId>
102
-
<artifactId>cuda</artifactId>
103
-
<version>10.0-7.3-1.4.3</version>
104
-
<classifier>macosx-x86_64-redist</classifier>
105
-
</dependency>
106
-
<dependency>
107
-
<groupId>org.bytedeco.javacpp-presets</groupId>
108
-
<artifactId>cuda</artifactId>
109
-
<version>10.0-7.3-1.4.3</version>
110
-
<classifier>windows-x86_64-redist</classifier>
73
+
<groupId>org.bytedeco</groupId>
74
+
<artifactId>cuda-platform-redist</artifactId>
75
+
<version>10.2-7.6-1.5.2</version>
111
76
</dependency>
112
77
113
78
Also note that, by default, Deeplearning4j will use the fastest algorithms available according to cuDNN, but memory usage may be excessive, causing strange launch errors. When this happens, try to reduce memory usage by using the [`NO_WORKSPACE` mode settable via the network configuration](/api/{{page.version}}/org/deeplearning4j/nn/conf/layers/ConvolutionLayer.Builder.html#cudnnAlgoMode-org.deeplearning4j.nn.conf.layers.ConvolutionLayer.AlgoMode-), instead of the default of `ConvolutionLayer.AlgoMode.PREFER_FASTEST`, for example:
0 commit comments