NVIDIA DGX Spark and Ollama Integration

Ollama and NVIDIA have partnered to ensure the NVIDIA DGX Spark runs local language models fast and efficiently out-of-the-box. This integration provides a high-performance environment for prototyping and running local LLMs using the Ollama framework.

Hardware Specifications of NVIDIA DGX Spark

The NVIDIA DGX Spark is powered by the NVIDIA GB10 Grace Blackwell Superchip, providing 1 petaFLOP of performance for local model execution. The system is equipped with 128GB of memory, enabling the support of large-scale models from providers such as Alibaba (Qwen), DeepSeek, DeepSeek, Meta (Llama), Mistral, Google (Gemma), and OpenAI (Gpt-oss).

Model Compatibility and Customization

Ollama enables users to run models from its extensive library or upload custom and fine-tuned models to the NVIDIA DGX Spark. The hardware's memory capacity allows for the support of various state-of-the-art models from the Ollama library.

Performance Optimization and Use Cases

Ollama is currently working with NVIDIA to optimize performance across several key AI workflows. These optimizations are focused on the following use cases:

• Chat: Standard conversational AI interfaces.
• Document Processing: Including retrieval, OCR, and modification.
• Document Processing: Including retrieval, OCR and modification.
• Code Tasks: Programming and software development assistance.
• Multimodal Workflows: Tasks involving multiple types of data inputs.