February 2024 Summaries

Fly.io has partnered with Tigris to provide an S3-compatible object storage solution for developers, making it easier and faster to handle file uploads in full-stack applications. The integration allows users to use the AWS libraries to put and delete objects into Tigris using the same calls that they use for S3. Fly.io has also wrapped everything under one bill, allowing users to pay a single monthly fee for their computer, block storage, databases, networking, and object storage needs.

Fly.io now offers GPU instances for users to leverage powerful computing capabilities such as large language models, text transcription, and image generation. The available GPU cards include Ampere A100 (40GB), Ampere A100 (80GB), and Lovelace L40s (48GB). Users can easily deploy a GPU instance by changing the `vm.size` for their apps or processes to any of these GPU kinds. Fly.io is also working on adding lower-cost A10 GPUs in the coming weeks.

Fly.io shares similar ergonomics as serverless platforms and aims to run some code on a fresh not-server when an event is received. By using the queue system, SQS, and their API for creating files on new Machines, events can be transformed into compute by triggering Fly Machines. The platform offers flexibility in choosing base images and server types per event while providing full control over Machine VM's responding to the events.

The text discusses a method of delegating resource-intensive tasks to more powerful machines only when necessary in web applications. It provides an example using Next.js, where the application is split into two process groups: one for accepting HTTP traffic and handling most requests (web), and another for performing heavy lifting (worker). The solution involves creating a route that performs the resource-intensive task, a wrapper function to delegate tasks between web and worker machines, and defining the process groups and Machines in the fly.toml configuration file. This method can improve application performance by offloading demanding tasks to more capable resources when needed.