NVIDIA DGX Spark

2025-10-12 · Source: Ollama Blog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure · Depth: Intermediate, quick

Summary

NVIDIA has released the DGX Spark, a new system powered by the NVIDIA GB10 Grace Blackwell Superchip, offering 1 petaFLOP of performance. This system is designed for prototyping and running local language models efficiently with Ollama, a partner in optimizing its out-of-the-box performance. Equipped with 128GB of memory, the DGX Spark supports a wide range of current models from major developers like Alibaba (Qwen), DeepSeek, Meta (Llama), Mistral, Google (Gemma), and OpenAI (Gpt-oss) via Ollama's library. Users can also upload custom or fine-tuned models. Ongoing optimization efforts between Ollama and NVIDIA focus on common use cases such as chat, document processing, code tasks, and multimodal workflows.

Key takeaway

For AI/ML Directors evaluating on-premise infrastructure for large language model development, the NVIDIA DGX Spark provides a powerful, optimized solution. Its 1 petaFLOP performance and 128GB memory, coupled with Ollama integration, enable rapid prototyping and execution of diverse models locally. Consider this system to accelerate your team's ability to experiment with and deploy custom or open-source LLMs without cloud dependencies.

Key insights

The NVIDIA DGX Spark offers 1 petaFLOP performance for local LLM prototyping via Ollama.

Principles

• Local LLM execution is critical for rapid iteration.
• Memory capacity dictates model size and complexity.

In practice

• Run large language models locally with 128GB memory.
• Utilize Ollama for diverse model access and custom uploads.

Topics

• NVIDIA DGX Spark
• Ollama
• Grace Blackwell Superchip
• Local Language Models
• Multimodal AI

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Engineer, Machine Learning Engineer, AI Architect

Related on AIssential

• NVIDIA DGX Spark performance — Ollama on NVIDIA DGX Spark delivers strong LLM inference performance across various models and quantizations.
• Run Local AI Agents with Faster Models and Multi-Node Clustering on NVIDIA DGX Spark — Integrated hardware and software solutions significantly accelerate local AI agent development and deployment, enhancing security and scalability.
• Lambda’s NVIDIA HGX 8xB200 on STAC-AI™ LANG6 — NVIDIA HGX 8xB200 significantly outperforms H200 NVL in LLM inference latency and throughput, especially under load and for large models.
• The $4,000 Backpack Supercomputer That Changed My Mind About Local AI — Unified memory architectures like NVIDIA's Grace Blackwell overcome VRAM limitations for local large language model deployment.
• Improved performance and model support with GGUF — Ollama 0.30 significantly boosts performance and broadens model/hardware compatibility via GGUF and llama.cpp integration.
• New Software and Model Optimizations Supercharge NVIDIA DGX Spark — NVIDIA's DGX Spark leverages NVFP4 and software optimizations for significant local AI model performance and memory efficiency.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Ollama Blog.