Azure Updates: AI infrastructure; Cost Management; Image Analysis

Product marketing manager Rachel Pruitt looked into the growing popularity of natural language processing, automated speech recognition, and text-to-speech. NVIDIA recently launched the open beta of the NVIDIA NeMo Megatron framework for building large language models. Microsoft is working to combine these NVIDIA capabilities with Azure AI infrastructure. Recently, Microsoft used Azure NDm A100 v4-series VMs to run GPT-3 models with the NeMo Megatron framework. Ultimately, Microsoft ran a 530B benchmark on 175 VMs with a training time per step of as little as 55.7 seconds. Pruitt wrote:

These results highlight an almost linear speed increase, guaranteeing better performance for a higher number of nodes—paramount for heavy or time-sensitive workloads. As shown by these runs with billions of parameters, customers can rest assured that Azure’s infrastructure can handle even the most difficult and complex workloads, on demand.

Principal product manager Michael Flanakin recapped the most recent Cost Management updates. Microsoft has introduced Azure savings plans allowing customers to save up to 65 percent on compute costs, ways to group costs by Virtual Desktop host pool, GA advisor scores, and cost optimization for Migrate.