TensorFlow 1.4.1 (GPU, CUDA 9)

GPU wheels for TensorFlow 1.4.1, built by TinyMind, the cloud machine learning platform.

To use the wheels on your own machine, Intel Broadwell or later CPU, and Nvidia computing capability 3.7 or later GPU with CUDA 9 and CuDNN 7 are required.

This version is optimized for compute capabilities 3.7 (K80, AWS P2/GCP), 6.0 (P100, GCP) and 7.0 (V100, AWS P3).

NOTE: These wheels contain MKL support. If you don't have it, install MKL by following the instructions here.

TensorFlow 1.4.1 (GPU)

GPU wheels for TensorFlow 1.4.1, built by TinyMind, the cloud machine learning platform.

To use the wheels on your own machine, Intel Broadwell or later CPU, and Nvidia computing capability 3.7 or later GPU with CUDA 8 and CuDNN 6 are required.

NOTE: Starting from TF 1.5, we will move to provide CUDA 9 wheels only. You should be able to upgrade your system to CUDA 9 and enjoy the performance boost. If you need CUDA 8 wheels for future versions, file an issue. If there is enough demand, we may consider keeping CUDA 8 builds around.

For TF 1.4.1, you can find CUDA 9 wheels here.

NOTE: These wheels contain MKL support. If you don't have it, install MKL by following the instructions here.

TensorFlow 1.4 (GPU, CUDA 9, Compute 3.7/6.0/7.0, no MKL)

GPU wheels for TensorFlow 1.4, built by TinyMind, the cloud machine learning platform.

To use the wheels on your own machine, Intel Broadwell or later CPU, and Nvidia computing capability 3.7 or later GPU with CUDA 9 and CuDNN 7 are required.

This version is optimized for compute capabilities 3.7 (K80, AWS P2/GCP), 6.0 (P100, GCP) and 7.0 (V100, AWS P3).

These wheels don't contain MKL support. It's recommended that you install MKL to your system to speed up TensorFlow unless you are unable to do so. Wheels with MKL support can be found at:

• CUDA 9 + Compute 3.7 if you have Nvidia K80 or equivalent GPU, or
• CUDA 9 + Compute 3.7, 6.0, 7.0 if you have Nvidia P100, V100 or equivalent GPU.

TensorFlow 1.4 (GPU, CUDA9, Generic)

CUDA 9 wheels for TensorFlow 1.4, without optimization. We provide them here as TensorFlow hasn't officially provided CUDA 9 wheels for TF 1.4, and our optimized builds may not work for AMD or old Intel CPUs.

If however you do have a relatively recent Intel CPU, you are looking for

• CUDA 9 + Compute 3.7 if you have Nvidia K80 or equivalent GPU, or
• CUDA 9 + Compute 3.7, 6.0, 7.0 if you have Nvidia P100, V100 or equivalent GPU.

TensorFlow 1.4 (CPU, macOS)

Optimized TF 1.4 wheels for macOS 10.12 (Sierra). We don't have plan to regularly build macOS wheels, but we will release them when we do.

These are built with Skylake CPUs (late 2016 MBP).

TensorFlow 1.4 (GPU, CUDA 9, Compute 3.7)

GPU wheels for TensorFlow 1.4, built by TinyMind, the cloud machine learning platform.

To use the wheels on your own machine, Intel Broadwell or later CPU, and Nvidia computing capability 3.7 or later GPU with CUDA 9 and CuDNN 7 are required.

If you are using Nvidia P100 or V100 GPU, you may want to use these optimized wheels instead.

NOTE: These wheels contain MKL support. If you don't have it, install MKL by following the instructions here.

TensorFlow 1.4 (GPU)

GPU wheels for TensorFlow 1.4, built by TinyMind, the cloud machine learning platform.

To use the wheels on your own machine, Intel Broadwell or later CPU, and Nvidia computing capability 3.7 or later GPU with CUDA 8 and CuDNN 6 are required.

NOTE: Starting from TF 1.5, we will move to provide CUDA 9 wheels only. You should be able to upgrade your system to CUDA 9 and enjoy the performance boost. If you need CUDA 8 wheels for future versions, file an issue. If there is enough demand, we may consider keeping CUDA 8 builds around.

For TF 1.4, you can find CUDA 9 wheels for K80 (Compute Capability 3.7) here, and for P100, V100 (Compute Capabilities 6.0, 7.0) here.

NOTE: These wheels contain MKL support. If you don't have it, install MKL by following the instructions here.

TensorFlow 1.4 (Debug)

These are builds for debugging TensorFlow itself. Here for performance-optimized builds? You are likely looking for this release.

TensorFlow 1.4 (GPU, CUDA 9, Compute 3.7/6.0/7.0)

GPU wheels for TensorFlow 1.4, built by TinyMind, the cloud machine learning platform.

To use the wheels on your own machine, Intel Broadwell or later CPU, and Nvidia computing capability 3.7 or later GPU with CUDA 9 and CuDNN 7 are required.

This version is optimized for compute capabilities 3.7 (K80, AWS P2/GCP), 6.0 (P100, GCP) and 7.0 (V100, AWS P3). If you are on K80, these wheels are built only with 3.7 and are smaller.

NOTE: These wheels contain MKL support. If you don't have it, install MKL by following the instructions here.

TensorFlow 1.4 (CPU)

CPU wheels for TensorFlow 1.4, built by TinyMind, the cloud machine learning platform.

To use the wheels on your own machine, Intel Broadwell or later CPU is required.