Fly.io has GPUs now (fly.io)

Article summary

Fly.io, a cloud provider, now offers GPUs for AI workloads, allowing users to run compute-intensive tasks near their users. The GPUs are available in several US and EU regions, as well as Sydney, with prices starting at $2.50/hr. Users can deploy their own code with favorite models on top of Fly.io's cloud backbone. The service also allows for scaling down to zero, only paying for GPU time when a user makes a request.

Main themes

• GPU availability
• Pricing and cost
• Edge computing
• Cloud providers comparison
• Scalability and performance
• Model loading and spin up times

What commenters say

• Some users are interested in the pricing model and how it compares to other cloud providers, with some finding it expensive and others seeing it as a good value.
• The speed of spin up and down, as well as model loading times, are important considerations for users evaluating the service.
• Comparisons to other cloud providers, such as AWS and DigitalOcean, reveal varying opinions on pricing and performance.
• Some users appreciate the simplicity and value for money offered by Fly.io, but others are deterred by the lack of certain features, such as a S3 replacement.
• The availability of GPUs and the ability to scale down to zero are seen as major advantages by some users, particularly for experimental or low-traffic projects.
• Concerns about latency and the speed of light limiting the effectiveness of edge computing are raised, but Fly.io's global presence is seen as a mitigating factor.
• The difficulty of obtaining GPUs and the resulting high demand are cited as reasons for the relatively high prices charged by Fly.io and other providers.