The latest ZenDNN Plugin for TensorFlow* (zentf) 5.2.0 is here!
The ZenDNN plugin for TensorFlow is called zentf.
The zentf 5.2.0 plugin works seamlessly with TensorFlow version 2.20.0, offering a high-performance experience for deep learning on AMD EPYC™ platforms.
We welcome feedback, suggestions, and bug reports. Should you have any of the these, please kindly file an issue on the ZenDNN Plugin for TensorFlow Github page: https://github.com/amd/ZenDNN-tensorflow-plugin/issues
AMD copyrighted code in ZenDNN is subject to the Apache-2.0, MIT, or BSD-3-Clause licenses; consult the source code file headers for the applicable license. Third party copyrighted code in ZenDNN is subject to the licenses set forth in the source code file headers of such code.
The following is a high-level block diagram for the zentf package which utilizes ZenDNN as the core inference library:
This file shows how to implement, build, install and run a TensorFlow-ZenDNN plug-in for AMD CPUs.
Refer to the support matrix for the list of supported operating system.
- Python
- Java
- C++
| Tools/Frameworks | Version |
|---|---|
| Bazel | 7.4.1 |
| Git | >=1.8 |
| Python | >=3.9 and <=3.13 |
| TensorFlow | 2.20.0 |
This section explains how to use the Python interface. For Java and C++ interfaces, kindly look inside the respective folders within the scripts folder.
- Create conda environment and activate it.
$ conda create -n tf-v2.20.0-zendnn-v5.2.0-rel-env python=3.10 -y $ conda activate tf-v2.20.0-zendnn-v5.2.0-rel-envNote: Python 3.10 used here for example.
- Install TensorFlow v2.20.0
$ pip install tensorflow==2.20.0
$ pip install zentf==5.2.0
-
Download the package and the user-guide from AMD developer portal.
-
Run the following commands to unzip the package and install the binary.
Note: We are taking an example for release package with Python version 3.10.
$ unzip ZENTF_v5.2.0_Python_v3.10.zip $ cd ZENTF_v5.2.0_Python_v3.10/ $ pip install zentf-5.2.0-cp310-cp310-manylinux_2_28_x86_64.whl -
To use the recommended environment settings, execute :
$ source scripts/zentf_env_setup.sh
$ git clone https://github.com/amd/ZenDNN-tensorflow-plugin.git
$ cd ZenDNN-tensorflow-plugin/
Note: Repository is defaults to main branch, to build the version 5.2.0 checkout the r5.2 branch.
$ git checkout r5.2
Notes:
export ZENDNNL_MANYLINUX_BUILD=1is needed for build from source for RHEL/FEDORA/Almalinux/CentOS OS families.- Configure & Build Tensorflow-ZenDNN Plug-in manually by following the steps [3-6].
The setup script will configure & build and install Tensorflow-ZenDNN Plug-in. It will also set the necessary environment variables of ZenDNN execution. However, these variables should be verified empirically.
ZenDNN-tensorflow-plugin$ source scripts/zentf_setup.sh
ZenDNN-tensorflow-plugin$ ./configure
You have bazel 7.4.1 installed.
Please specify the location of python. [Default is /home/user/anaconda3/envs/zentf-env/bin/python]:
Found possible Python library paths:
/home/user/anaconda3/envs/zentf-env/lib/python3.10/site-packages
Please input the desired Python library path to use. Default is [/home/user/anaconda3/envs/zentf-env/lib/python3.10/site-packages]
Do you wish to build TensorFlow plug-in with MPI support? [y/N]:
No MPI support will be enabled for TensorFlow plug-in.
Please specify optimization flags to use during compilation when bazel option "--config=opt" is specified [Default is -march=native -Wno-sign-compare]:
Configuration finished
ZenDNN-tensorflow-plugin$ bazel clean --expunge
ZenDNN-tensorflow-plugin$ bazel build -c opt //tensorflow_plugin/tools/pip_package:build_pip_package --verbose_failures --copt=-Wall --copt=-Werror --spawn_strategy=standalone
ZenDNN-tensorflow-plugin$ bazel-bin/tensorflow_plugin/tools/pip_package/build_pip_package .
Note: It will generate and save python wheel file for TensorFlow-ZenDNN Plug-in into the current directory (i.e., ZenDNN-tensorflow-plugin/).
ZenDNN-tensorflow-plugin$ pip install zentf-5.2.0-cp310-cp310-linux_x86_64.whl
The build and installation from source is done!
$ export TF_ENABLE_ZENDNN_OPTS=1
$ export TF_ENABLE_ONEDNN_OPTS=0
Note: To disable ZenDNN optimizations in your inference execution, you can set the corresponding ZenDNN environment variable
export TF_ENABLE_ZENDNN_OPTS=0
ZenDNN-tensorflow-plugin$ python tests/softmax.py
2026-02-10 03:41:43.885189: I external/local_xla/xla/tsl/cuda/cudart_stub.cc:31] Could not find cuda drivers on your machine, GPU will not be used.
2026-02-10 03:41:43.885871: I tensorflow/core/util/port.cc:180] ZenDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable `TF_ENABLE_ZENDNN_OPTS=0`.
2026-02-10 03:41:43.945721: I tensorflow/core/platform/cpu_feature_guard.cc:210] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
To enable the following instructions: AVX2 AVX512F AVX512_VNNI AVX512_BF16 FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.
2026-02-10 03:41:46.251297: I tensorflow/core/util/port.cc:180] ZenDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable `TF_ENABLE_ZENDNN_OPTS=0`.
2026-02-10 03:41:46.253122: I external/local_xla/xla/tsl/cuda/cudart_stub.cc:31] Could not find cuda drivers on your machine, GPU will not be used.
2026-02-10 03:41:46.525058: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303)
Tensor("random_normal:0", shape=(10,), dtype=float32)
2026-02-10 03:41:47.004884: I tensorflow/core/common_runtime/direct_session.cc:381] Device mapping: no known devices.
WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
I0000 00:00:1770720107.005847 2394608 mlir_graph_optimization_pass.cc:437] MLIR V1 optimization pass is not enabled
random_normal/RandomStandardNormal: (RandomStandardNormal): /job:localhost/replica:0/task:0/device:CPU:0
2026-02-10 03:41:47.009820: I tensorflow/core/common_runtime/placer.cc:162] random_normal/RandomStandardNormal: (RandomStandardNormal): /job:localhost/replica:0/task:0/device:CPU:0
random_normal/mul: (Mul): /job:localhost/replica:0/task:0/device:CPU:0
2026-02-10 03:41:47.009846: I tensorflow/core/common_runtime/placer.cc:162] random_normal/mul: (Mul): /job:localhost/replica:0/task:0/device:CPU:0
random_normal: (AddV2): /job:localhost/replica:0/task:0/device:CPU:0
2026-02-10 03:41:47.009857: I tensorflow/core/common_runtime/placer.cc:162] random_normal: (AddV2): /job:localhost/replica:0/task:0/device:CPU:0
Softmax: (Softmax): /job:localhost/replica:0/task:0/device:CPU:0
2026-02-10 03:41:47.009867: I tensorflow/core/common_runtime/placer.cc:162] Softmax: (Softmax): /job:localhost/replica:0/task:0/device:CPU:0
random_normal/shape: (Const): /job:localhost/replica:0/task:0/device:CPU:0
2026-02-10 03:41:47.009878: I tensorflow/core/common_runtime/placer.cc:162] random_normal/shape: (Const): /job:localhost/replica:0/task:0/device:CPU:0
random_normal/mean: (Const): /job:localhost/replica:0/task:0/device:CPU:0
2026-02-10 03:41:47.009885: I tensorflow/core/common_runtime/placer.cc:162] random_normal/mean: (Const): /job:localhost/replica:0/task:0/device:CPU:0
random_normal/stddev: (Const): /job:localhost/replica:0/task:0/device:CPU:0
2026-02-10 03:41:47.009892: I tensorflow/core/common_runtime/placer.cc:162] random_normal/stddev: (Const): /job:localhost/replica:0/task:0/device:CPU:0
2026-02-10 03:41:47.010443: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:117] Plugin optimizer for device_type CPU is enabled.
[0.05660784 0.09040404 0.03201076 0.11204024 0.2344563 0.162052
0.09466095 0.11205972 0.0752109 0.03049729]
- zentf v5.2.0 is supported with ZenDNN v5.2.0. For detailed performance tuning guidelines, refer to the Performance Tuning section of the ZenDNN user guide.
To check the version of zentf use the following command:
python -c 'import zentf; print(zentf.__version__)'To check the build config of zentf use the following command:
python -c 'import zentf; print(*zentf.__config__.split("\n"), sep="\n")'