Existing Confirmed

Paper on XGLM

Meta AI • December 20, 2021
Total GPUs
256
H100 Equiv
80
Power
0.2 MW
FP16
80 PF

System Equivalents

This datacenter's 256 GPUs equal:

NVIDIA DGX A100
8 GPUs/system
32
NVIDIA HGX A100 (8-GPU)
8 GPUs/system
32

Specifications

Primary GPUNVIDIA A100
GPU Count256
SectorPrivate

Additional Information

"The XGLM 7.5B model was trained on 256 A100 GPUs for about 3 weeks, at a speed of 311.6k words per second"