GPU Architecture: NVIDIA Pascal

NVIDIA CUDA Core: 3584

Double-Precision Performance: 5.3 TeraFLOPS

Single-Precision Performance: 10.6 TeraFLOPS

Half-Precision Performance: 21.2 TeraFLOPS

GPU Memory: 16 GB CoWoS HBM2

Memory Bandwidth: 732 GB/s

Interconnect: NVIDIA NVLink

Max Power Consumption: 300 W

ECC: Native support with no capacity or performance overhead

Thermal Solution: Passive

Form Factor: SXM2

Compute APIs: NVIDIA CUDA, DirectCompute, OpenCL, OpenACC