Hi, folks ~
So this week, I've been trying my best to get a version of TensorFlow up and running such that it can talk to my GPU. I've typed out a rough description of the steps I've taken below the equals signs.
I've got a pretty low-end AMD GPU that came with a CPU I bought back in 2013 (integrated graphics). These are:
CPU: AMD A10-6800K APU (4) @ 4.1GHz
GPU: AMD ATI Radeon HD 8670D
I found out that I needed ROCm, an open source high performance computing (HPC) platform developed by AMD. But my first concern is that my CPU and GPU don't show up on the list of supported hardware here (if I am reading correctly): https://towardsdatascience.com/train-neural-networks-using-amd-gpus-and-keras-37189c453878
When I do the tests mentioned on the following page (/opt/rocm/bin/rocminfo && /opt/rocm/opencl/bin/clinfo):
Only rocminfo prints my GPU info, the other program doesn't exist. Is it something I did wrong during the install? Or just the fact that my hardware (maybe) isn't supported? There seems to be some indication that it may run on unsupported hardware anyway....
Thanks for reading, any help would be appreciated.
==Rough Steps, sourced primarily from FAQ - Installation Instructions for ArchLinux - AMD Community ==
- Installed yay
- Ran yay -S rocm-dkms
- Ran yay -S tensorflow-rocm
- Installed pyenv so that I could switch into python 3.8 to use TF.
- Once in the pyenv shell, I ran pip3 install tensorflow-rocm
- This is the session from tensorflow, indicating to me that it does not talk to the GPU (could not retrieve ROCM device count):
import tensorflow as tf
2021-06-16 16:04:21.183521: I tensorflow/compiler/jit/xla_cpu_device.cc:41] Not creating XLA devices, tf_xla_enable_xla_devices not set
2021-06-16 16:04:21.183827: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library libamdhip64.so
2021-06-16 16:04:21.203053: E tensorflow/stream_executor/rocm/rocm_driver.cc:982] could not retrieve ROCM device count: HIP_ERROR_NoDevice
2021-06-16 16:04:21.203489: I tensorflow/core/platform/cpu_feature_guard.cc:142] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations: SSE3 SSE4.1 SSE4.2 AVX FMA
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
2021-06-16 16:04:21.205371: I tensorflow/compiler/jit/xla_gpu_device.cc:99] Not creating XLA devices, tf_xla_enable_xla_devices not set
2021-06-16 16:04:21.205475: E tensorflow/stream_executor/rocm/rocm_driver.cc:982] could not retrieve ROCM device count: HIP_ERROR_NoDevice