Existing Confirmed

Paper on Megatron-Turing

Microsoft,NVIDIA • January 28, 2022
Total GPUs
4.5K
H100 Equiv
1.4K
Power
3.8 MW
FP16
1.4K PF

System Equivalents

This datacenter's 4.5K GPUs equal:

NVIDIA DGX A100
8 GPUs/system
560
NVIDIA HGX A100 (8-GPU)
8 GPUs/system
560

Specifications

Primary GPUNVIDIA A100 SXM4 80 GB
GPU Count4.5K
SectorPrivate

Additional Information

"Model training is done with mixed precision using 16-bit bfloat on NVIDIA’s Selene [2] supercomputer with 560 DGX A100 nodes. Each cluster node has 8 NVIDIA 80-GB A100 GPUs"