Azure Container Apps is a serverless container running service that charges based on actual usage. It offers the option to leverage specialized chips for running AI workloads, with the choice of NVIDIA T4 and A100 chips.
Even though it is already operational, regular customers who have not agreed to a Microsoft Enterprise Agreement must contact Microsoft for serverless GPU quota before usage. Currently, it is available in two data centers: West US 3 and Australia East.
For developers working on LLM applications, Microsoft has introduced Azure Container Apps Dynamic Sessions in this round. This service limits container operations to a sandbox environment to receive code from LLM and run it. The running containers are isolated within Hyper-V from the external world. This service was launched in May but has now entered GA status for the Python Interpreter and has added a preview for Node.js.
Various features announced at Ignite, such as Private Endpoints restricting access only from Azure Virtual Network, Planned Maintenance for scheduling update periods, and Path-Based Routing to separate apps based on URL paths without the need for a reverse proxy.
Source: App on Azure Blog
TLDR: Azure Container Apps offers serverless container running with specialized chips for AI workloads, requires contacting Microsoft for serverless GPU quota, and introduces Dynamic Sessions for limited container operations in a sandbox environment. Various additional features like Private Endpoints and Path-Based Routing have been announced.
Leave a Comment