hipDNN installation

hipDNN installation#

2026-03-31

4 min read time

Applies to Linux and Windows

The hipDNN frontend API is distributed as a header-only library. It requires a ROCm development package installation. These development packages contain the .dev prefix.

System requirements#

An AMD GPU with ROCm support
Linux or Windows operating system (see Install AMD ROCm)

Dependencies#

Dependency	Version	Description
ROCm development installation	ROCm 7.12.0	AMD GPU programming stack
CMake	3.25.2 or later	Build system generator
Ninja (recommended)	1.12.1 or later	Faster build system
C++ Compiler	C++17 compatible	hipDNN API supports C++17

Install ROCm and hipDNN#

Follow the instructions at Install AMD ROCm to install a ROCm development package that includes hipDNN. ROCm development packages contain the .dev prefix.

Important

The base ROCm install package doesn’t include the hipDNN frontend API header files. You must use the development ROCm packages.

For example, on Linux with an AMD Instinct MI350X Series GPU, install the amdrocm-core-dev7.12-gfx950 package instead of the amdrocm7.12-gfx950 package (amdrocm-core-dev7.12-gfx950 includes the amdrocm7.12-gfx950 package, so only the amdrocm-core-dev7.12-gfx950 package needs to be installed).

Verify hipDNN installation#

The hipDNN library ships with executables that can be used to test the hipDNN library and plugins when the GPU hardware is available. These executables can be run individually or as a suite using the ctest driver program. (ctest is installed as part of the CMake package.)

Run hipDNN unit tests#

A suite of unit test executables are included with hipDNN. These test executables don’t require the GPU hardware to run.

Here’s an example that demonstrates running these tests using ctest:

$ ctest --test-dir /opt/rocm/bin/hipdnn
Internal ctest changing into directory: /opt/rocm/bin/hipdnn
Test project /opt/rocm/bin/hipdnn
    Start 1: hipdnn_data_sdk_tests
1/7 Test #1: hipdnn_data_sdk_tests ............   Passed    0.27 sec
    Start 2: hipdnn_backend_tests
2/7 Test #2: hipdnn_backend_tests .............   Passed    1.25 sec
    Start 3: hipdnn_frontend_tests
3/7 Test #3: hipdnn_frontend_tests ............   Passed    0.04 sec
    Start 4: hipdnn_test_sdk_tests
4/7 Test #4: hipdnn_test_sdk_tests ............   Passed    3.48 sec
    Start 5: hipdnn_plugin_sdk_tests
5/7 Test #5: hipdnn_plugin_sdk_tests ..........   Passed    0.03 sec
    Start 6: public_hipdnn_backend_tests
6/7 Test #6: public_hipdnn_backend_tests ......   Passed    0.27 sec
    Start 7: public_hipdnn_frontend_tests
7/7 Test #7: public_hipdnn_frontend_tests .....   Passed    0.23 sec

100% tests passed, 0 tests failed out of 7

Replace /opt/rocm with your ROCm install folder when running this ctest command.

Run hipDNN samples#

The sample programs shipped with hipDNN require AMD GPU hardware to execute. The samples execute supported tensor operations on GPU hardware using the plugins installed with hipDNN.

This example demonstrates running the entire set of samples using ctest:

$ ctest --test-dir /opt/rocm/bin/hipdnn_samples
Internal ctest changing into directory: /opt/rocm/bin/hipdnn_samples
Test project /opt/rocm/bin/hipdnn_samples
      Start  1: conv_fprop
1/15 Test  #1: conv_fprop .............................   Passed    5.15 sec
      Start  2: conv_dgrad
2/15 Test  #2: conv_dgrad .............................   Passed    5.80 sec
      Start  3: conv_wgrad
3/15 Test  #3: conv_wgrad .............................   Passed    8.03 sec
      Start  4: fused_conv_fprop_activ
4/15 Test  #4: fused_conv_fprop_activ .................   Passed    1.01 sec
      Start  5: fused_conv_fprop_bias_activ
5/15 Test  #5: fused_conv_fprop_bias_activ ............   Passed    1.43 sec
      Start  6: conv_fprop_deterministic
6/15 Test  #6: conv_fprop_deterministic ...............   Passed    9.14 sec
      Start  7: bn_inference
7/15 Test  #7: bn_inference ...........................   Passed    2.78 sec
      Start  8: bn_inference_with_variance
8/15 Test  #8: bn_inference_with_variance .............   Passed    0.45 sec
      Start  9: bn_training
9/15 Test  #9: bn_training ............................   Passed    3.26 sec
      Start 10: fused_bn_training_activ
10/15 Test #10: fused_bn_training_activ ................   Passed    3.30 sec
      Start 11: bn_backward
11/15 Test #11: bn_backward ............................   Passed    3.52 sec
      Start 12: fused_bn_inference_activ
12/15 Test #12: fused_bn_inference_activ ...............   Passed    2.81 sec
      Start 13: fused_bn_inference_drelu_bn_backward
13/15 Test #13: fused_bn_inference_drelu_bn_backward ...   Passed    3.08 sec
      Start 14: fused_bn_inference_variance_activ
14/15 Test #14: fused_bn_inference_variance_activ ......   Passed    0.48 sec
      Start 15: serialization_roundtrip
15/15 Test #15: serialization_roundtrip ................   Passed    6.49 sec

100% tests passed, 0 tests failed out of 15