Can a Compute Capability 3.0 card run Tensorflow 1.8 tensorflow-gpu runtime?

Question

Going through the install tutorial for Linux (Tensorflow 1.8) and I'm not sure how to interpret the phrase:

GPU card with CUDA Compute Capability 3.0 or higher for building from source and 3.5 or higher for our binaries. See NVIDIA documentation for a list of supported GPU cards.

I have an NVIDIA GTX 770 which has Compute Capability 3.0, does that mean I can build the tensorflow-gpu binary but I can't use it to run/execute tensorflow-gpu? (current CUDA version for TF is 9.0 + cuDNN 7.x)

Answer 1

No, it means that if you have a compute capability 3.0 card, you have to build and install tensorflow-gpu from the sources, you can't use the pre-built binaries to install with pip.

It will have full functionality once installed.

This is because there are a lot of different code variants for different compute capability cards, so to keep the binary to a reasonable size, only a selected range of compute capabilities are supported in the pre-built binary.