Chainer 3.4.0 と mnist

前回、imagenet では性能差が出た２つの GPU ですが、mnist に関しては差は無いです（何故？）。。。

◯計算サーバ
CPU : Intel(R) Xeon(R) CPU E5-2687W v4 @ 3.00GHz x 2
メモリ：512GB
GPU : NIVIDIA Tesla P100 x 2
OS : CentOS 7.4

$ time python ./train_mnist.py -g 0
GPU: 0
# unit: 1000
# Minibatch-size: 100
# epoch: 20

epoch main/loss validation/main/loss main/accuracy validation/main/accuracy elapsed_time
1 0.193723 0.0904885 0.940934 0.9725 4.87579
2 0.0743769 0.0801725 0.976116 0.9751 7.75057
3 0.0489755 0.0797656 0.984049 0.9764 11.0034
4 0.0343368 0.0808019 0.989065 0.9773 14.0952
5 0.0291952 0.0724612 0.989865 0.98 16.9548
6 0.0242752 0.0770883 0.992065 0.9789 20.1595
7 0.0205797 0.0871283 0.993248 0.978 23.1138
8 0.0179716 0.0806064 0.994265 0.98 26.4275
9 0.0156324 0.0726476 0.994965 0.9831 29.4012
10 0.0158118 0.0868025 0.995015 0.981 32.4502
11 0.0120386 0.0940075 0.996282 0.9795 35.4172
12 0.0146274 0.0938886 0.995532 0.981 38.3322
13 0.011219 0.093875 0.996699 0.9815 41.5736
14 0.0109149 0.0940998 0.996516 0.981 44.4308
15 0.00991011 0.105109 0.997265 0.9803 47.2963
16 0.0142973 0.0860043 0.995499 0.9823 50.1539
17 0.00960365 0.113148 0.997232 0.9789 53.0158
18 0.00765447 0.104115 0.997649 0.982 55.8894
19 0.0100436 0.103909 0.997065 0.9803 59.0706
20 0.00832253 0.0919329 0.997749 0.9828 62.1928

real 1m6.997s
user 1m6.568s
sys 0m14.379s


$ time python ./train_mnist.py -g 1
GPU: 1
# unit: 1000
# Minibatch-size: 100
# epoch: 20

epoch main/loss validation/main/loss main/accuracy validation/main/accuracy elapsed_time
1 0.190071 0.110005 0.943967 0.9656 3.44792
2 0.0737473 0.088046 0.977165 0.972 6.72184
3 0.0466676 0.096217 0.984783 0.9717 9.752
4 0.0366625 0.0668992 0.987815 0.98 12.8739
5 0.0285734 0.081525 0.990698 0.979 15.9065
6 0.0242173 0.0797278 0.992031 0.9797 19.4017
7 0.0212822 0.0786974 0.993082 0.9793 22.6291
8 0.0183767 0.0908352 0.994115 0.98 25.4699
9 0.0191029 0.100031 0.994232 0.9763 28.6368
10 0.0127762 0.0779863 0.996149 0.9823 31.7296
11 0.0143983 0.085073 0.995132 0.9813 34.9406
12 0.0107437 0.098093 0.996816 0.9779 38.213
13 0.0153186 0.0922849 0.995749 0.9822 41.3357
14 0.0101606 0.0987587 0.996565 0.9814 44.4927
15 0.0124978 0.0775459 0.996249 0.983 47.8019
16 0.00857302 0.0966257 0.997199 0.9816 50.9471
17 0.0106839 0.093041 0.997216 0.9836 54.0331
18 0.00887633 0.109647 0.997249 0.9805 57.2198
19 0.0117919 0.112135 0.996682 0.9809 60.3072
20 0.00785438 0.0926466 0.997766 0.9831 63.1354

real 1m7.586s
user 1m8.398s
sys 0m13.939s