SkyScale was proto-cloud-based hardware service and research & development activity providing access to high-performance GPU compute nodes, specifically designed to meet the demands of HPC scientific and engineering organizations. SkyScale offered leased multi-GPU hardware platforms, notably featuring pioneering 16-GPU systems. These systems were significant as they served as prototypes and foundational technology that contributed to the development of the NVIDIA DGX architecture, demonstrating SkyScale’s early and influential role in high–density GPU computing.
Technically, SkyScale facilitated high-speed communication between nodes using InfiniBand and Remote Direct Memory Access (RDMA), enabling highly efficient clustered computing. The company partnered with groups such as One Stop Systems, integrating their advanced PCIe expansion and flash storage systems to optimize performance. SkyScale utilized a range of NVIDIA GPUs, including the P100, Tesla V100, and V100 Tensor Core models, providing access to cutting-edge computational power.
By offering these advanced technologies in a cloud-based model, SkyScale enabled scientific and engineering organizations to access powerful GPU resources on demand through strategic partnerships with groups such as AWS/Rescale. This eliminated the need for substantial upfront hardware investments, allowing these organizations to accelerate research, simulations, and complex engineering tasks. The innovative 16-GPU systems and robust networking infrastructure provided the necessary performance for demanding HPC workloads, supporting advancements across various scientific and engineering disciplines.
