OllamaFrom the builder
NVIDIA DGX Spark
NVIDIA DGX Spark is now available, and Ollama worked with NVIDIA so Ollama runs fast and efficiently on it out of the box. Powered by the GB10 Grace Blackwell Superchip, it delivers 1 petaFLOP of performance and 128GB of memory, enough for models such as Qwen, DeepSeek, Llama, Mistral, Gemma and gpt-oss. Ollama says it is continuing to optimize performance with NVIDIA across chat, document processing, coding and multimodal workflows.
You can also upload and bring your own custom or fine-tuned models.
Saved . A shortened excerpt from the official blog; the original publisher retains authorship.