NVIDIA to Offer DGX Spark With 64GB Memory Option
NVIDIA plans to release a new version of its DGX Spark system this month, this time with 64GB of unified memory. Unified memory lets a computer's processor and other chips share the same memory pool, which helps when running demanding AI workloads locally instead of relying on the cloud. According to NVIDIA Blog, local AI is growing more useful as AI agents, which are programs that can act on tasks with less human input, move from experiments into daily development work. Open models, whose underlying code is publicly available, are also shrinking in size, letting them run on more devices. The bigger memory option gives developers more room to build and scale AI projects directly on their own hardware.
Words to know
- DGX Spark
- — NVIDIA's computer system designed for building and running AI workloads
- unified memory
- — A shared memory pool that different chips in a device can access together
- AI agents
- — AI programs that can act on tasks with less direct human guidance
- open models
- — AI models whose code is publicly available for anyone to use or modify
- local AI
- — Running AI models directly on a device instead of through the cloud
Summary written from NVIDIA Blog's headline and teaser