1. Introduction

GPUs are very expensive resources

Purchasing GPUs for on-premise use or allocating them in cloud environments is notoriously expensive, and securing high-performance GPUs in the cloud can be quite challenging.

Unless you're developing computationally intensive applications like LLMs, you're probably spending most of your time writing or debugging code. The actual GPU runtime likely accounts for only a small fraction of the entire development process.

This led us to a key realization: let's use GPUs only when they're truly needed.