yueyericardo
commented
3 years ago Current result on TITAN V
python tools/aev-benchmark-size.py
File: small.pdb, Molecule size: 264
/home/richard/dev/torchani_cuaev/torchani/resources/
Downloading ANI model parameters ...
Original TorchANI:
Duration: 1.26 s
Speed: 2.53 ms/it
CUaev:
Duration: 0.25 s
Speed: 0.51 ms/it
Speed up: 4.97 X
----------------------------------------------------------------------
File: 1hz5.pdb, Molecule size: 973
/home/richard/dev/torchani_cuaev/torchani/resources/
Original TorchANI:
Duration: 1.21 s
Speed: 2.43 ms/it
CUaev:
Duration: 2.42 s
Speed: 4.83 ms/it
Speed up (slower): 0.50 X
----------------------------------------------------------------------
File: 6W8H.pdb, Molecule size: 3410
/home/richard/dev/torchani_cuaev/torchani/resources/
Original TorchANI:
Duration: 1.86 s
Speed: 3.72 ms/it
CUaev:
Duration: 27.69 s
Speed: 55.38 ms/it
Speed up (slower): 0.07 Xpython tools/training-aev-benchmark.py /home/richard/dev/ANI-1x-wb97xdz.h
=> loading dataset...
=> loading /home/richard/dev/ANI-1x-wb97xdz.h5, total molecules: 3114
3114/3114 [==============================] - 73.4s
=> Caching shuffled dataset...
=> loading /home/richard/dev/ANI-1x-wb97xdz.h5, total molecules: 3114
3114/3114 [==============================] - 66.5s
=> CUDA info:
Total devices: 1
0: TITAN V
_CudaDeviceProperties(name='TITAN V', major=7, minor=0, total_memory=12036MB, multi_processor_count=80)
GPU Memory Cached (pytorch) : 0.0MB / 12036.7MB (TITAN V)
GPU Memory Used (nvidia-smi): 11.9MB / 12036.7MB (TITAN V)
=> Test 1: USE cuda extension, Energy training
=> start training
Epoch: 1/1
1935/1935 [========] - 32s 16ms/step - rmse: 55.5933
GPU Memory Cached (pytorch) : 826.0MB / 12036.7MB (TITAN V)
GPU Memory Used (nvidia-smi): 2088.9MB / 12036.7MB (TITAN V)
=> More detail about benchmark PER EPOCH
Total AEV - 3.892 sec
Forward - 7.081 sec
Backward - 8.709 sec
Force - 0.0 ms
Optimizer - 8.386 sec
Others - 3.525 sec
Epoch time - 31.593 sec
=> Test 2: NO cuda extension, Energy training
=> start training
Epoch: 1/1
1935/1935 [========] - 50s 26ms/step - rmse: 61.1153
GPU Memory Cached (pytorch) : 2278.0MB / 12036.7MB (TITAN V)
GPU Memory Used (nvidia-smi): 3540.9MB / 12036.7MB (TITAN V)
=> More detail about benchmark PER EPOCH
Total AEV - 23.749 sec
Forward - 6.979 sec
Backward - 8.809 sec
Force - 0.0 ms
Optimizer - 8.274 sec
Others - 2.393 sec
Epoch time - 50.204 sec
=> Test 3: USE cuda extension, Force and Energy inference
=> start training
Epoch: 1/1
1935/1935 [========] - 43s 22ms/step - rmse: 1316.6297
GPU Memory Cached (pytorch) : 2972.0MB / 12036.7MB (TITAN V)
GPU Memory Used (nvidia-smi): 4234.9MB / 12036.7MB (TITAN V)
=> More detail about benchmark PER EPOCH
Total AEV - 3.926 sec
Forward - 7.434 sec
Backward - 0.0 ms
Force - 22.671 sec
Optimizer - 0.0 ms
Others - 8.975 sec
Epoch time - 43.006 sec
=> Test 4: NO cuda extension, Force and Energy inference
=> start training
Epoch: 1/1
1935/1935 [========] - 88s 46ms/step - rmse: 21.6047
GPU Memory Cached (pytorch) : 10054.0MB / 12036.7MB (TITAN V)
GPU Memory Used (nvidia-smi): 11316.9MB / 12036.7MB (TITAN V)
=> More detail about benchmark PER EPOCH
Total AEV - 25.259 sec
Forward - 7.305 sec
Backward - 0.0 ms
Force - 48.022 sec
Optimizer - 0.0 ms
Others - 7.652 sec
Epoch time - 88.239 sec
lgtm-com[bot]
This pull request introduces 1 alert when merging 32910c45a50d00cc780549b1533ac6267f5a85f6 into 23c9816c5d6490ac4e0fe80b98bf78b5be10cef5 - view on LGTM.com
new alerts: