NVIDIA DGX Spark
The NVIDIA DGX Spark platform is set to lead the charge in a paradigm shift for AI development. Designed for developers, data scientists, and AI researchers, this system enables secure and local AI. It delivers petaflop-scale performance and empowers local AI development with scalability and advanced features – all in a compact form that fits in the palm of your hand. The Spark is ideal for startups or small businesses that lack the necessary cloud resources or any AI developer looking to avoid unnecessary network exposure. It is also a great educational tool for organizations offering programs that teach AI programming.
Don’t let its humble appearance deceive you, the Spark packs quite the processing punch. Powered by the NVIDIA GB10 superchip, it combines their 20 core Grace CPU and AI Blackwell GPU to optimize data preprocessing and orchestration, accelerate model tuning and enable real time inference. 128 GB of LPDDR5X unified system memory provides the bandwidth and capacity needed for smooth model development, rapid experimentation, and high efficiency.
The GB10 Superchip combines two major innovative pieces of NVIDIA technology: The Grace CPU and Blackwell GPU. NVIDIA Grace CPUs use high-performance Arm Neoverse V2 cores coupled with NVIDIA Scalable Coherency Fabric. NVIDIA Blackwell GPUs pack 208 billion transistors and features two reticle-limited dies connected by a 10 terabytes per second (TB/s) chip-to-chip interconnect in a unified single GPU.
Working together as the GB10 Superchip, this enables serious technological breakthroughs. As an SoC, it is capable of 1 petaflop of AI performance at FP4 precision. This is a level of computational power that in previous generations, would require an entire room’s worth of advanced hardware to accomplish.
Memory is key for developing AI. The Spark uses 128 GB of LPDDR5X memory with error-correction code for server-class reliability, and high energy efficiency. This is a class of memory specifically designed for high performance and efficiency and, critical for the DGX Spark, it allows both the CPU and GPU to access the memory space, further enhancing its capability.
Delivering a blazing fast 1000 TOPS, with the capability to run and scale LLMs locally for up to 200B parameters. NVIDIA ConnectX technology further enhances this capacity by allowing two platforms to combine for even further sophistication of up to 400B parameters. This means users will be able to develop LLMs securely and locally at a large scale without the need for unnecessary exposure or cloud resources.
These features come together to form an AI developer’s dream: A compact but powerful platform that, at 6x6x2 inches, fits in the palm of your hand and enables testing and development of AI models securely, right at your desk. The Spark provides a seamlessly scalable platform for ML ops, developing robotics, smart city technology, large scale models and simulations, and other computer vision solutions.
NVIDIA Spark Use Cases
Development and ML Ops
DGX Spark provides AI engineers and developers a seamlessly scalable platform for AI inferencing, training, and development locally and securely.
Education and Research
DGX Spark is a cost-effective system for universities and colleges to use in academic programs teaching students how to develop and program AI.
Media and Creative Industries
The Spark can be used for generative content creation such as image generation and natural language processing.
Varieties
ASI has Spark solutions available from multiple manufacturers, including MSI, ASUS, and PNY.
Contact your ASI sales rep or sales@asipartner.com and we will work with you to determine which option is best for you. For further product information, register below.