19 GPU nodes, each with single V100-PCIE-32GB
>> torch.cuda.device_count() >> 1
c[1612-1615] = 4 nodes
c[1701-1715] = 15 nodes
19 GPU nodes, each with single V100-PCIE-32GB
>> torch.cuda.device_count() >> 1
c[1612-1615] = 4 nodes
c[1701-1715] = 15 nodes