ResNet-50 batch_size=64 8卡時可以跑得起來,但是1/2/4卡時均跑不起來,顯示:
F0824 10:07:26.011203 35974 gpu_memory.hpp:38] Failed to allocate 40140800bytes on device 0. Total memory: 24025956352, Free: 33488896, dev_info[0]:total=24025956352 free=33488896
why?
ResNet-50 batch_size=64 8卡時可以跑得起來,但是1/2/4卡時均跑不起來,顯示:
F0824 10:07:26.011203 35974 gpu_memory.hpp:38] Failed to allocate 40140800bytes on device 0. Total memory: 24025956352, Free: 33488896, dev_info[0]:total=24025956352 free=33488896
why?