Data science teams are at the forefront of Artificial Intelligence (AI) innovation, developing projects that transform organisations and the products they create. IoT innovation requires high-performance computing resources capable of training complex IoT models.
Dedicated and fully optimised IoT platforms enable IoT teams to have a high-performance computing environment without investing heavily in high-performance computing (HPC) infrastructure – service stations, storage, low-latency networking, cooling and air-conditioning solutions etc. A standard power socket and an internet cable are sufficient.
NVIDIA DGX Station A100 – The platform for data science teams allows them to have the architecture and technology used by data centres right under their desk, without any additional IT infrastructure or data centre. Four NVIDIA A100 Tensor Core GPU accelerators feature 160GB of HBM2 or 320GB of HBM2e memory and a combined bandwidth of 2.4 TB/s, enabling data scientists to achieve performance equivalent to data centre technology anywhere, anytime.
NVIDIA A100 GPU accelerators are combined into a GPU cluster using third-generation NVLink technology to achieve bi-directional speeds of 300 GB/s, while the other devices (NVMe array and CPU) are served by PCIe Gen4 x16 infrastructure, generating 252 Gb/s.
Training workloads for IoT models and neural networks typically involve reading the same data sets multiple times to increase accuracy.
In addition, NVIDIA A100 GPU accelerators use a new segmentation technology called Multi-Instance GPU (MIG), which allows a single GPU accelerator to be divided into seven (7) logically separated GPU resources, meaning that up to twenty-eight (28) independent tasks of different types can be assigned to the NVIDIA DGX Station A100 server simultaneously, enabling the platform to use its resources as effectively as possible.
Such a platform has been purchased from 3RTechnology UAB by one of the largest universities in Estonia. The platform is exclusively dedicated to university data science teams working on research and practical application of IoT-based projects. The NVIDIA DGX A100 Station will serve as the main R&D (research and development) platform for the development, testing, training and evaluation of the IoT concepts themselves.
UAB 3RTechnology is an official NVIDIA Enterpise partner in the Baltic States with Compute DGX specialisation and certified systems engineers.
UAB 3RTechnology specializes in the design, consulting and implementation of platforms for Artificial Intelligence (AI) and Machine Learning (ML). Specialization includes preparation of High-Performance Computing (HPC) architectures, evaluation of potential computing cluster performance, selection and optimization of CPU and GPU resources using best practices and taking into account the client’s specific needs.