mikegerber/test-nvidia
1
0
Fork 0
Code Issues Pull requests Releases Wiki Activity
Test Nvidia environment in relation to TensorFlow
16 commits 1 branch 0 tags 38 KiB
Find a file
Gerber, Mike 4dbd9fbeea 📝 README: Results
2020-09-10 12:01:25 +02:00
2019-12-tensorflow2-keras-cnn+lstm 🧹 2019-12-tensorflow2-keras-cnn+lstm 2020-05-15 13:22:43 +02:00
2020-03-tensorflow-vs-tensorflow-gpu 🧹 2020-03-tensorflow-vs-tensorflow-gpu/tensorflow-vs-tensorflow-gpu 2020-05-15 13:32:27 +02:00
assets ✨ Test TensorFlow 2 2020-09-05 12:59:40 +02:00
README.md 📝 README: Results 2020-09-10 12:01:25 +02:00
run ✨ Test TensorFlow 2 2020-09-05 12:59:40 +02:00
run-docker ✨ Add run-docker-compatibility-matrix to test more combinations 2020-09-09 14:07:27 +02:00
run-docker-compatibility-matrix ✨ Add run-docker-compatibility-matrix to test more combinations 2020-09-09 14:07:27 +02:00
test-nvidia ✨ Add run-docker-compatibility-matrix to test more combinations 2020-09-09 14:07:27 +02:00

README.md

Test Nvidia CUDA environment in relation to TensorFlow

  • ./run tests the native system. One of tf1 or tf2 is expected to have no GPU available due to CUDA library incompatibility
  • ./run-docker tests Docker support. Both TensorFlow versions should work as we're using a base image compatible to the respective version.
  • ./run-docker-compatibility-matrix tests combinations of (pip-installable) TensorFlow versions and nvidia/cuda images.

Example output

% ./run-docker
== tf1
GPU 0: GeForce RTX 2080 (UUID: GPU-612ce75c-1340-772b-039c-2a83a3ea5c95)
TensorFlow 1.15.3
GPU available: True
== tf2
GPU 0: GeForce RTX 2080 (UUID: GPU-612ce75c-1340-772b-039c-2a83a3ea5c95)
TensorFlow 2.3.0
GPU available: True

Results

As of 2020-09, the only combinations that are working:

  • TensorFlow 1.15.3 using CUDA Toolkit 10.0
  • TensorFlow 2.3.0 using CUDA Toolkit 10.1

This is only for pip-installable TensorFlow, not self-compiled nor Anaconda. We also did not test other TensorFlow versions.