Nvidia Enables Linking of DGX Spark Systems for Expanded AI Capacity
Nvidia is now allowing users to connect up to four of its DGX Spark AI supercomputers, creating a system with a combined 512GB of shared random access memory. This development expands the capabilities of the DGX Spark, launched in early 2024, and addresses the growing demand for powerful AI infrastructure.
The DGX Spark, powered by the NVIDIA GB10 Grace Blackwell Superchip, delivers up to one petaFLOP of FP4 AI performance in a power-efficient package. The ability to link multiple units exponentially increases these capabilities, opening new possibilities for complex AI workloads, according to the company.
Scalability is achieved through NVIDIA’s NVLink-C2C interconnect and ConnectX-7 networking technology. NVLink provides a high-bandwidth, low-latency connection between the Grace CPU and the Blackwell GPU within each DGX Spark unit, as well as between multiple linked units. This coherent memory access aims to eliminate performance bottlenecks traditionally associated with PCIe connections.
ConnectX-7 further enhances data transfer speed and efficiency between the systems. The GB10 Grace Blackwell Superchip is a key component in this expanded capability. This move underscores a trend toward democratizing access to high-performance computing, extending beyond large data centers to developers and researchers.
Nvidia similarly offers resources for performance tuning of the DGX Spark, covering areas such as CPU and memory optimization, GPU power limits, storage, and Spark/RAPIDS configurations. A repository for optimized ComfyUI installation on the DGX Spark – both natively and via Docker – is available, with reported GPU performance improvements, as discussed in developer forums.
