NVIDIA released DGX Spark 64GB on October 23, saying it runs capable local agents on device without cloud dependency. It is the company's first local AI platform update since the launch of DGX Spark.
NVIDIA reported the new SKU supports up to 100-billion-parameter models, measured on local device performance. That compares with the 128GB model’s support for up to 200 billion parameters.
DGX Spark 64GB is built on NVIDIA Grace Blackwell compute and targets developers, researchers and AI enthusiasts. Availability begins on October 23, initially for manufacturer partners.
"DGX Spark combines NVIDIA Grace Blackwell compute, unified memory, NVIDIA ConnectX-7 networking and an NVIDIA CUDA-accelerated AI software stack in one system," said NVIDIA. The system is a complete local AI platform for agents, inference, fine-tuning, data science and edge development.
The announcement follows NVIDIA’s focus on local AI solutions. NVIDIA said the release expands access to high-performance computing for local AI development.
NVIDIA did not say when the 128GB model will be available, and raised the question of scaling for larger workloads. The company said the platform will support scaling as workloads demand.
Source: nvidia
