Thunder Compute’s mission is to make the best way to access GPUs. To do this, we created a way to virtualize GPUs over a network.
Up to 5x cheaper
GPU virtualization allows us to efficiently share GPUs so that you never pay for idle time
Scale from one environment
Scale from a CPU to a cluster of GPUs with one command
Simplified UX
Start coding in under 60 seconds. No config, code changes, or drivers required
Meet the Team
Built For Developers by Developers
Carl Peterson and Brian Model were close friends at Georgia Tech. With backgrounds in data science and systems architecture, both of us understood the need for better ways to access GPUs. Our vision for Thunder Compute is to make GPUs dramatically more efficient through virtualization, like how Amazon EBS virtualized storage and VMWare virtualized x86 processors.
At Thunder Compute, the greatest asset is the dedicated team, led by cloud computing experts Carl Peterson and Brian Model. Each brings a unique set of skills and experiences that drive the mission to deliver cutting-edge cloud solutions.
Innovative Approach
From Concept to Reality: Thunder's Journey
Brian had the idea for Thunder Compute when his research lab was using an excel sheet to sign up for GPU time two weeks in advance. This inspired us to begin brainstorming ways to efficiently allocate GPUs. We realized that the flexibility of GPU cloud systems is constrained by the physical connection between instances and GPUs. If it was possible to decouple each CPU from its GPU, we could build a system that shares GPUs extremely efficiently, giving developers the flexibility to instantly change their hardware.
1
The Problem
We hated developing with GPUs
Finding, configuring, and paying for GPUs is painful. No platform has created one-size-fits-all solution to simplify GPU development.
2
The Idea
Find a way to virtualize GPUs over a network
Virtualizing GPUs at the CUDA level separates the GPU hardware from its environment, allowing GPU sharing and flexibility to an unmatched level
3
Building
It took years to prove this worked
This problem was hard and we didn't know if it was even possible. It took years of trial and error to find an approach that worked.
4
Success
Seeing the vision come to life
Running a model with a GPU on a different computer for the first time was like magic. Nothing makes us smile more than hearing devs say "wow, it's really one line of code?"
Investors
Backed by Industry Leaders
We are extremely grateful to have partnerships with some of the strongest names in tech. Our partners make it possible to create the best experience in cloud for our developers. With their collaboration we will continue to push the bounds of what is possible, redefining what developers expect to be able to do with GPUs.