Sparxity, formerly known as SpiNNcloud, a German company specializing in neuromorphic supercomputing, has unveiled its sparse-native computing rack system at the Jülich Supercomputing Centre in Germany. The costly installation will facilitate the execution of sparse machine learning models and extensive high-performance computing simulations.
In an announcement, Sparxity revealed that its fresh brand image embodies its mission to process exclusively the essential information. As worries about energy consumption and costs persist in conventional AI infrastructure deployments, the company explained that sparse computation decreases the amount of computing and data that need to be moved, thereby reducing bottlenecks.
The JSC system exemplifies this approach, as it computes only when an event occurs. The cluster, called Jubis, was designed to provide tenfold faster sparse solver execution, 18 times lower energy consumption for event-driven ML workloads, and 20 times shorter time to reach an acceptable solution in optimization problems.
Once operational, the system will be accessible for broader high-performance computing (HPC) and artificial intelligence (AI) purposes. Hector Gonzalez, co-founder, and CEO of Sparxity stated, “Model builders have invested years creating workarounds for the constraints of compact hardware.” “Sparxity is designed to unleash the potential of its users by calculating and focusing on the most important aspects, thus overcoming any constraints.
The rebranding process we are undergoing places a significant emphasis on the sparse-native method, which serves as a functional basis for the development of the upcoming AI infrastructure. Our existing systems introduce scarcity to both HPC and AI, and our partnership with the Jülich Supercomputing Centre, which is among the largest supercomputing establishments in Europe, demonstrates the worth of that methodology on a supercomputer level.
Additional information can be found in the category of IT Hardware and Semiconductors.
