Thanks to Nvidia you no longer need millions (or even billions) of dollars of institutional backing to work on AI projects. The wildly successful chipmaker unveiled a tiny desktop device called Project Digits at this year’s convention that’s potent enough for the complex operations required to run and train AI models. What Digits lacks in size it makes up for in power, thanks to its GB100 “superchip," 128 gigabytes of high-bandwidth unified memory, and 4 terabytes of NVMe storage. It will start at $3,000, which is a lot less than the cost of the supercomputers being used by OpenAI and the rest of its peers.
From the article by Bryan Hood
