The next generation of Team Green’s GPUs
Following the launch of the Turing consumer GPU architecture in 2018, NVIDIA is getting ready for the next generation with Ampere. To debut with the A100 enterprise GPU, Ampere aims to deliver up to 6x better performance than NVIDIA’s Volta’s architecture. Designed for data centers, scientific computing, cloud graphics, and data analytics, the A100 is a 7nm-based GPU with over 54 billion transistors, making it one of the biggest 7nm-based processors out there.
The A100 comes with 19.5 teraflops of FP32 performance, 6,912 CUDA cores, 40GB of memory, and 1.6TB/s of memory bandwidth.
Together with the A100, NVIDIA is introducing the DGX A100. As the world’s first 5-petaflops server, the DGX A100 has eight A100 GPUs combined into one massive unit. With this layout, the DGX A100 comes with 320GB of GPU memory and a massive 12.4TB/s of memory bandwidth. In addition, the DGX A100 has 15TB of NVMe SSDs to store key AI training tasks.
With a starting price of $199,000(~Php 10.1m), the DGX A100 is meant for professional applications. There’s no word yet if NVIDIA will use Ampere architecture with its consumer GPUs.