aiqm / torchani

Accurate Neural Network Potential on PyTorch
https://aiqm.github.io/torchani/
MIT License
459 stars 126 forks source link

[WIP] cuaev calculation optimzation #577

Closed yueyericardo closed 9 months ago

yueyericardo commented 3 years ago

original benchmark result:

-----------------------------------------------------------------------------------------------------------------------------------------
RUN                 PDB          Size         forward      backward     Others       Total        Total(200)   Speedup      GPU
-----------------------------------------------------------------------------------------------------------------------------------------
01 py aev fd        small.pdb    264          1.8 ms       0.0 ms       0.0 ms       1.8 ms       368.1 ms     -            757.1MB
02 cu aev fd        small.pdb    264          0.5 ms       0.0 ms       0.0 ms       0.5 ms       106.9 ms     3.44         751.1MB
03 py aev fd        1hz5.pdb     973          2.7 ms       0.0 ms       0.0 ms       2.7 ms       535.9 ms     -            809.1MB
04 cu aev fd        1hz5.pdb     973          1.0 ms       0.0 ms       0.0 ms       1.0 ms       209.3 ms     2.56         771.1MB
05 py aev fd        6W8H.pdb     3410         7.0 ms       0.0 ms       0.0 ms       7.0 ms       1.394 sec    -            1121.1MB
06 cu aev fd        6W8H.pdb     3410         4.8 ms       0.0 ms       0.0 ms       4.8 ms       958.4 ms     1.45         1065.1MB
07 py aev fd        1C17.pdb     6000         24.5 ms      0.0 ms       0.0 ms       24.6 ms      4.912 sec    -            2113.1MB
08 cu aev fd        1C17.pdb     6000         15.5 ms      0.0 ms       0.0 ms       15.5 ms      3.108 sec    1.58         1761.1MB
09 py aev fd        1C17.pdb     10000        56.4 ms      0.0 ms       0.0 ms       56.4 ms      11.277 sec   -            3983.1MB
10 cu aev fd        1C17.pdb     10000        39.4 ms      0.0 ms       0.0 ms       39.4 ms      7.879 sec    1.43         3597.1MB
-----------------------------------------------------------------------------------------------------------------------------------------
11 py aev fd+bd     small.pdb    264          3.8 ms       3.5 ms       0.0 ms       7.3 ms       1.457 sec    -            811.1MB
12 cu aev fd+bd     small.pdb    264          0.6 ms       0.8 ms       0.0 ms       1.4 ms       282.6 ms     5.16         811.1MB
13 py aev fd+bd     1hz5.pdb     973          3.9 ms       4.3 ms       0.0 ms       8.2 ms       1.648 sec    -            1047.1MB
14 cu aev fd+bd     1hz5.pdb     973          1.1 ms       1.3 ms       0.0 ms       2.4 ms       489.2 ms     3.37         1023.1MB
15 py aev fd+bd     6W8H.pdb     3410         8.0 ms       4.2 ms       0.0 ms       12.2 ms      2.440 sec    -            1365.1MB
16 cu aev fd+bd     6W8H.pdb     3410         4.9 ms       1.3 ms       0.0 ms       6.2 ms       1.241 sec    1.97         1631.1MB
17 py aev fd+bd     1C17.pdb     6000         24.8 ms      21.3 ms      0.0 ms       46.1 ms      9.225 sec    -            3869.1MB
18 cu aev fd+bd     1C17.pdb     6000         15.7 ms      6.9 ms       0.0 ms       22.6 ms      4.521 sec    2.04         4667.1MB
19 py aev fd+bd     1C17.pdb     10000        -            -            -            -            -            -            -
20 cu aev fd+bd     1C17.pdb     10000        39.9 ms      11.1 ms      0.0 ms       51.0 ms      10.199 sec   -            5415.1MB
-----------------------------------------------------------------------------------------------------------------------------------------

new benchmark result:


-----------------------------------------------------------------------------------------------------------------------------------------
RUN                 PDB          Size         forward      backward     Others       Total        Total(200)   Speedup      GPU
-----------------------------------------------------------------------------------------------------------------------------------------
01 py aev fd        small.pdb    264          1.9 ms       0.0 ms       0.0 ms       2.0 ms       394.9 ms     -            757.1MB
02 cu aev fd        small.pdb    264          0.6 ms       0.0 ms       0.0 ms       0.6 ms       122.9 ms     3.21         749.1MB
03 py aev fd        1hz5.pdb     973          2.8 ms       0.0 ms       0.0 ms       2.8 ms       555.3 ms     -            809.1MB
04 cu aev fd        1hz5.pdb     973          0.8 ms       0.0 ms       0.0 ms       0.8 ms       161.7 ms     3.43         763.1MB
05 py aev fd        6W8H.pdb     3410         7.1 ms       0.0 ms       0.0 ms       7.1 ms       1.415 sec    -            1121.1MB
06 cu aev fd        6W8H.pdb     3410         1.1 ms       0.0 ms       0.0 ms       1.1 ms       215.4 ms     6.57         975.1MB
07 py aev fd        1C17.pdb     6000         24.8 ms      0.0 ms       0.0 ms       24.8 ms      4.956 sec    -            2113.1MB
08 cu aev fd        1C17.pdb     6000         3.6 ms       0.0 ms       0.0 ms       3.6 ms       721.2 ms     6.87         1487.1MB
09 py aev fd        1C17.pdb     10000        56.7 ms      0.0 ms       0.0 ms       56.7 ms      11.346 sec   -            3983.1MB
10 cu aev fd        1C17.pdb     10000        6.1 ms       0.0 ms       0.0 ms       6.1 ms       1.216 sec    9.33         2831.1MB
-----------------------------------------------------------------------------------------------------------------------------------------
11 py aev fd+bd     small.pdb    264          3.8 ms       3.6 ms       0.0 ms       7.4 ms       1.480 sec    -            811.1MB
12 cu aev fd+bd     small.pdb    264          0.6 ms       0.9 ms       0.0 ms       1.5 ms       301.5 ms     4.91         807.1MB
13 py aev fd+bd     1hz5.pdb     973          4.1 ms       4.4 ms       0.0 ms       8.5 ms       1.698 sec    -            1047.1MB
14 cu aev fd+bd     1hz5.pdb     973          0.9 ms       1.3 ms       0.0 ms       2.2 ms       442.1 ms     3.84         1015.1MB
15 py aev fd+bd     6W8H.pdb     3410         8.2 ms       4.3 ms       0.0 ms       12.6 ms      2.513 sec    -            1365.1MB
16 cu aev fd+bd     6W8H.pdb     3410         1.1 ms       1.3 ms       0.0 ms       2.5 ms       500.3 ms     5.02         1251.1MB
17 py aev fd+bd     1C17.pdb     6000         25.0 ms      21.3 ms      0.0 ms       46.4 ms      9.277 sec    -            3869.1MB
18 cu aev fd+bd     1C17.pdb     6000         3.7 ms       6.7 ms       0.0 ms       10.4 ms      2.075 sec    4.47         3797.1MB
19 py aev fd+bd     1C17.pdb     10000        -            -            -            -            -            -            -
20 cu aev fd+bd     1C17.pdb     10000        6.2 ms       10.8 ms      0.0 ms       17.0 ms      3.397 sec    -            3843.1MB
-----------------------------------------------------------------------------------------------------------------------------------------
yueyericardo commented 3 years ago

According to the notice, following updates are moved privately.