***Note: this benchmark is out of date. ***
| Name | Value |
|---|---|
| CPU | E5-2680 v2 |
| GPU | K40m |
| Memory | DDR3, 256GB @ 1600Mhz |
| Name | Value |
|---|---|
| OS | Ubuntu 14.04 LTS |
| GPU | driver 367.44 |
| CUDA | 7.5 |
| cuDNN | V5.1 |
| Name | Batch Size | Input | Output | Layers | #Parameters |
|---|---|---|---|---|---|
| FCN-5 | 8192 | 512 | 1000 | 5 | ~11 millions |
| FCN-8 | 8192 | 512 | 1000 | 8 | ~24millions |
| AlexNet | 16 | 150528 | 1000 | 4 | ~62 millions |
| ResNet-50 | 16 | 150528 | 1000 | 50 | ~25 millions |
| LSTM-32 | 128 | 10000 | 10000 | 2 | ~6 millions |
| LSTM-64 | 128 | 10000 | 10000 | 2 | ~6 millions |
| Tool | Version |
|---|---|
| Caffe | 4ba654f |
| CNTK | v1.7.1 |
| MXNet | 32cb6bc |
| TensorFlow | r0.10 |
| Theano | 0.8.X |
| Torch | 95f137f |
https://github.com/n17s/hkbu-benchmark
https://github.com/FreemanX/hkbu-benchmark
Disclaimer: I'm a Microsoft employee, however, this is my personal github account and information/code shared here does not represent opinions or views of Microsoft in any way.