Week In Review: Auto, Security, Pervasive Computing
Edge, cloud, data center The NVIDIA’s new A100 GPU is targeting data centers and AI at the same time. The partitioning feature on the 7nm processor will enable data centers to slice the workloads and keep the servers busy. Based on NVIDIA’s Ampere architecture with sparsity for AI, the GPU also has third generation Tensor Cores with TF32 that NVIDIA says speeds up AI training and inference by up to 20x. The GPU has seven independent instances for inferencing tasks and NVLink interconnect to enable putting multiple GPUs together. The chip is in full production and is now shipping in the NVIDIA DGX A100, a 5-petaflops server.