Benchmarks for Deep Learning models implemented in TensorFlow. These benchmarks are easy to reproduce if you already have TensorFlow installed on your machine. If you have a different hardware, feel free to contribute.
For more in-depth benchmarks, see:
Timing benchmark for AlexNet inference. For more details refer to the paper ImageNet Classification with Deep Convolutional Neural Networks.
Running the benchmark:
python tensorflow/models/image/alexnet/alexnet_benchmark.py
| Num | GPU/CPU | Memory | Forward pass (ms) | Forward-backward pass (ms) | Details |
|---|---|---|---|---|---|
| 1 | Titan X | 12GB GDDR5 | 70 +/- 0.1 | 244 +/- 30 | as reported in alexnet_benchmark.py |
| 2 | GeForce GTX 960M | 2GB | 121 +/- 0 | 359 +/- 1 | Dell XPS 15 9550 / Ubuntu 16.04 / CUDA v7.5 / cuDNN 5.1 |
| 3 | K40c | 12GB GDDR5 | 145 +/- 1.5 | 480 +/- 48 | as reported in alexnet_benchmark.py |
| 4 | GeForce GT 750M | 2GB | 536 +/- 2 | 1466 +/- 18 | MacBook Pro Late 2013 / OS X 10.11.6 / CUDA v7.5 / cuDNN 5.1 |
| 5 | 2.3 GHz Intel Core i7 | 16 GB 1600 MHz DDR3 | 2473 +/- 34 | 7091 +/- 117 | MacBook Pro Late 2013 / OS X 10.11.6 / CUDA v7.5 / cuDNN 5.1 |
Benchmark for image recognition using a convolutional neural network (CNN) in CIFAR-10 dataset. For more details refer to the TensorFlow tutorial.
Running the benchmark:
python tensorflow/models/image/cifar10/cifar10_train.py
| Num | GPU/CPU | Memory | Examples/second | Seconds/batch | Details |
|---|---|---|---|---|---|
| 1 | GTX 1080 | 8 GB GDDR5X | 1780.0 | 0.072 | as reported in tf benchmarks |
| 2 | GeForce GTX 960M | 2GB | 1529 +/- 68 | 0.0839 +/- 0.0041 | Dell XPS 15 9550 / Ubuntu 16.04 / CUDA v7.5 / cuDNN 5.1 |
| 3 | Titan X | 12GB GDDR5 | 550.1 | 0.233 | as reported in tf benchmarks |
| 4 | GeForce GT 750M | 2GB | 535.33 +/- 15 | 0.2393 +/- 0.007 | MacBook Pro Late 2013 / OS X 10.11.6 / CUDA v7.5 / cuDNN 5.1 |
| 5 | 2.3 GHz Intel Core i7 | 16 GB 1600 MHz DDR3 | 336.6 +/- 15 | 0.38 +/- 0.015 | MacBook Pro Late 2013 / OS X 10.11.6 / CUDA v7.5 / cuDNN 5.1 |