NVIDIA Supercomputer

The NVIDIA DGX™ A100 supercomputer is a universal system for all AI workloads—from analytics to training to inference. It is the first off-the-shelf hardware/software stack specifically designed for Deep Neural Networks (DNN) workloads for artificial intelligence (AI) applications.

The DGX™ A100 is the first compute platform which delivers a 5 PetaFLOPS AI Performance in a single node.  In terms of today’s practical computing speed, the petaflop can also be thought of as one million gigaflops, where the gigaflop represents 1 billion floating-point operations per second where commonly available machines have not yet reached this standard.

All this hardware comes pre-loaded with NVIDIA’s software stack, including a Deep Learning (DL) SDK that optimizes DL computations for single- and multi-GPU processing and supports all major DL programming frameworks through NVIDIA-Docker, a GPU-enabled version of the popular Docker containerization engine.

Typical Applications:

• Reliability engineering, safety analysis and simulation
• Safety-critical autonomous systems on ground and air

Specifications:

• AI performance: 5 petaFLOPS in a Single Node
• Inference Performance: 10 PetaOPS
• Training performance (NLP: BERT-Large) 1200 Seq/s
• 8x Core GPUs in one system