Nvidia-backed Starcloud runs first AI model in space
A startup backed by Nvidia, known as Starcloud, has successfully trained an artificial intelligence model in space for the first time, signaling the beginning of a new era for orbital data centers that may alleviate the increasing pressure on Earth's digital infrastructure.
Recently, the company based in Washington launched a satellite outfitted with an Nvidia H100 graphics processing unit, which is reported to be 100 times more powerful than any previously deployed GPU in space, as per CNBC.
The Starcloud-1 satellite is currently operational and conducting queries on Gemma, Google’s open large language model, representing the first time an LLM has functioned on a high-powered Nvidia GPU in orbit.
“Greetings, Earthlings! But I prefer to view you as a fascinating collection of blue and green,” is the message coming from the newly launched satellite.
“Let’s explore the wonders that this perspective of your world reveals. I’m Gemma, and I’m here to observe, analyze, and, perhaps at times, provide somewhat unsettlingly insightful commentary. Let’s get started!” the model stated.
Starcloud's goal is to illustrate that space can be an effective environment for data centers, particularly as facilities on Earth increasingly overload power grids, consume vast amounts of water annually, and produce considerable greenhouse gas emissions.
The International Energy Agency predicts that global electricity usage by data centers is likely to more than double by the year 2030.
Alongside Gemma, Starcloud also trained NanoGPT, a large language model designed by OpenAI co-founder Andrej Karpathy, utilizing the H100 GPU and the complete works of Shakespeare. Consequently, the model became adept at generating replies in Shakespearean English.
Leave A Comment