HomeCompaniesThunder Compute

Thunder Compute

Virtualization software to make GPUs 5x cheaper

GPUs are in extremely high demand for AI development. A single GPU server can cost up to $300K. Despite this, GPUs in data centers are only utilized an average of 15-20% of the time. Thunder Compute is building technology to increase GPU utilization from 15-20% to 90%+, meaning that each GPU can be shared by 5x as many developers as in data centers today. Our technology achieves this efficiency gain by allowing all of the servers in a data center to share a single pool of network-attached GPUs rather than physically attaching a GPU to each server. Try Thunder Compute at docs.thundercompute.com/quickstart

Jobs at Thunder Compute

Atlanta, GA, US
$90K - $110K
0.50% - 1.00%
Any (new grads ok)
Thunder Compute
Founded:2024
Team Size:3
Location:Atlanta, GA
Group Partner:Aaron Epstein

Active Founders

Carl Peterson, Founder

CEO of Thunder Compute, formerly Management Consultant at Bain
Carl Peterson
Carl Peterson
Thunder Compute

Brian Model, Founder

CTO of Thunder Compute, formerly Quantitative Developer at Citadel Securities

Company Launches

With Thunder Compute, run your current workload on cloud GPUs with one line of code. You can develop without paying for a GPU and then instantly connect your environment to as many GPUs as you need to move to production.

Watch it in action

https://www.youtube.com/watch?v=qsOBFQZtsFM

Hi everyone, we’re Carl and Brian, aka the team behind Thunder Compute.

Before Thunder Compute, Carl was a Management Consultant at Bain & Company, and Brian was a Quantitative Developer at Citadel Securities. Carl and Brian met at Georgia Tech and have been close friends for over 6 years.

Problem

Developers spend hours configuring cloud platforms to meet the ever-changing needs of their projects. Today’s cloud platforms trade-off between:

  • Scalability
  • Cost
  • Simplicity

Solution

Thunder Compute is a flexible cloud environment that adapts to the needs of any project with a single command, for example:

Using your local filesystem?

$ tnr start --sync

Scaling to 50 A100 GPUs?

$ tnr device A100 -n 50

Running a Jupyter Notebook?

$ notebook

Developing without a GPU?

$ tnr device cpu

All of this is available at industry-leading prices and can be set up in minutes - just pip install tnr.

Our innovation

Think VMware for GPUs: Thunder Compute tricks your computer into thinking it’s attached to a GPU, which actually sits across a network. This means that CPU-only machines can behave exactly as if they have dedicated GPUs, while the physical GPUs are actually shared among several machines.

Our offer

To help us test our systems at scale, we are giving free open beta access to NVIDIA T4 instances. Try Thunder Compute today at thundercompute.com. all we ask in return is that you please report any bugs 🙂

If you have questions, need technical support, or want to rant about the current state of GPUs, join our discord at https://discord.gg/nwuETS9jJK or contact us at founders@thundercompute.com.

YC Sign Photo

YC Sign Photo

Hear from the founders

How did your company get started? (i.e., How did the founders meet? How did you come up with the idea? How did you decide to be a founder?)

Carl Peterson and Brian Model met as freshmen at Georgia Tech and became close friends. The idea for Thunder Compute sparked from Brian's experience in his Systems for AI lab, where researchers made GPU reservations weeks in advance via Google Sheets, severely hindering research progress. This inefficiency inspired us to explore ways to improve GPU utilization and the developer experience of using GPUs. \Our thesis was (and still is) that virtualization over a network is the optimal way to manage GPUs within a cloud platform and founded Thunder Compute to bring our technology to market.

YC S24 Application Video