How to run TensorFlow on AMD/ATI GPU?

6
votes

After reading this tutorial https://www.tensorflow.org/guide/using_gpu I checked GPU session on this simple code

import numpy as np
import matplotlib.pyplot as plt
import tensorflow as tf

a = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[2,3], name = 'a')
b = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape = [3,2], name =  'b')
c = tf.matmul(a, b)

with tf.Session(config=tf.ConfigProto(log_device_placement=True)) as sess:
    x = sess.run(c)
print(x)

The output was

2018-08-07 18:44:59.019144: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA Device mapping: no known devices. 2018-08-07 18:44:59.019536: I tensorflow/core/common_runtime/direct_session.cc:288] Device mapping:

MatMul: (MatMul): /job:localhost/replica:0/task:0/device:CPU:0 2018-08-07 18:44:59.019902: I tensorflow/core/common_runtime/placer.cc:886] MatMul: (MatMul)/job:localhost/replica:0/task:0/device:CPU:0 a: (Const): /job:localhost/replica:0/task:0/device:CPU:0 2018-08-07 18:44:59.019926: I tensorflow/core/common_runtime/placer.cc:886] a: (Const)/job:localhost/replica:0/task:0/device:CPU:0 b: (Const): /job:localhost/replica:0/task:0/device:CPU:0 2018-08-07 18:44:59.019934: I tensorflow/core/common_runtime/placer.cc:886] b: (Const)/job:localhost/replica:0/task:0/device:CPU:0 [[ 22. 28.] [ 49. 64.]]

As you see there is no calculation done by GPU. and when I changed the code to use GPU's configuration and process fraction:

conf = tf.ConfigProto()
conf.gpu_options.per_process_gpu_memory_fraction = 0.4

with tf.Session(config = conf) as sess:
    x = sess.run(c)
print(x)

The output was

2018-08-07 18:52:22.681221: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA [[ 22. 28.] [ 49. 64.]]

What can I do to run the session on GPU card? Thank you.

3
tensorflowgpuamd

3 Answers

0
votes

I believe TensorFlow-GPU only support GPU card with CUDA Compute Capability >= 3.0 of NVIDIA.

The following TensorFlow variants are available for installation:

TensorFlow with CPU support only. If your system does not have a NVIDIA® GPU, you must install this version. This version of TensorFlow is usually easier to install, so even if you have an NVIDIA GPU, we recommend installing this version first.

TensorFlow with GPU support. TensorFlow programs usually run much faster on a GPU instead of a CPU. If you run performance-critical applications and your system has an NVIDIA® GPU that meets the prerequisites, you should install this version. See TensorFlow GPU support for details.

https://www.tensorflow.org/install/install_linux

5
votes

It is most certainly possible to run tensorflow on AMD GPUs. About 2 years back ROCm was released which gets things done. However, the is a caveat, that it runs only on Linux as of now owing to its open-source origins. So if you are willing to use Linux then you can most certainly train your DL models using AMD GPUs. That said the amount of support you will get is low as the community is still not large enough. Google search for ROCm and you can get instructions on how to get it set up and running on a Linux machine. May be it will work with WSL2 in windows, but I have not tried it yet and so cannot comment on that.

here is a link to ROCm installation docs

4
votes

You can use TensorflowJS, the Javascript version of tensorflow. TensorflowJS does not have any HW limitation and can run on all the gpu supporting webGL.

The api is pretty similar to tf in python and the project provides scripts to convert your models from python to JS