Existing Confirmed

Paper on Minibatch SGD

Meta AI • United States of America • June 8, 2017
Total GPUs
256
H100 Equiv
2
Power
0.2 MW
FP16
5 PF

Specifications

Primary GPUNVIDIA P100
GPU Count256
SectorPrivate

Additional Information

The training time of ResNet-101 is 92.5 minutes in our implementation using 256 Tesla P100 GPUs and a minibatch size of 8k