Hi,
I have two computers: with GPU GTX 1050 and TITAN X. I start training the same network… and GTX 1050 is faster than TITAN X. Why?
First with GTX 1050:
Epoch 1/1
200/200 [==============================] - 338s 2s/step - loss: 1.8443 - rpn_class_loss: 0.0644 - rpn_bbox_loss: 0.4029 - mrcnn_class_loss: 0.3557 - mrcnn_bbox_loss: 0.5603 - mrcnn_mask_loss: 0.4610 - val_loss: 2.6352 - val_rpn_class_loss: 0.0606 - val_rpn_bbox_loss: 1.1116 - val_mrcnn_class_loss: 0.3390 - val_mrcnn_bbox_loss: 0.6298 - val_mrcnn_mask_loss: 0.4942
CPU times: user 5min 41s, sys: 15 s, total: 5min 56s
Wall time: 6min 1s
And second with TITAN X:
Epoch 1/1
200/200 [==============================] - 688s 3s/step - loss: 1.7756 - rpn_class_loss: 0.0602 - rpn_bbox_loss: 0.4342 - mrcnn_class_loss: 0.2609 - mrcnn_bbox_loss: 0.5811 - mrcnn_mask_loss: 0.4392 - val_loss: 1.5796 - val_rpn_class_loss: 0.0519 - val_rpn_bbox_loss: 0.4217 - val_mrcnn_class_loss: 0.2369 - val_mrcnn_bbox_loss: 0.4825 - val_mrcnn_mask_loss: 0.3867
CPU times: user 8min 32s, sys: 16.9 s, total: 8min 49s
Wall time: 11min 46s