NVIDIA has just made its new Ampere GPU architecture official, unveiling the first Ampere-based GPU in the A100 GPU. You can read more about that here, but alongside the NVIDIA A100 GPU is the new DGX A100 system.
NVIDIA's new DGX A100 3rd Generation integrated AI system packs a huge amount of computing power, centered around 8 x NVIDIA A100 GPUs, as well as 2 x 64-core/128-thread AMD Rome CPUs with 1TB of RAM. There's a huge 15TB of PCIe 4.0 NVMe SSDs for kick ass storage, and 9 x Mellanox ConnectX-6 VPI for 200Gb/s over the network.
The new DGX A100 is built for data science and AI, with up to 5 PFLOPs of peak FP16 performance, and up to 2.5 TFLOPs of TF32 performance, right up to 156 TFLOPs with FP64 performance.