School of AI in a 985 university
Meeting the computing power needs of everyday and urgent projects
Cooperation background: When new students joined the research group, it is necessary to purchase GPU servers according to the number of people to support the experiment, which consumes a lot of energy in hardware procurement, storage, and management; Due to project cycles and other reasons, there were insufficient resources available during short-term GPU computing power peaks.
Cooperation benefits: PARATERA Smart Computing Cloud collaborated deeply with the school, utilizing cloud resources and cloud accounts to flexibly meet the diverse GPU computing power needs of account changes and permission allocation generated by the project team member turnover, and short-term and large amounts of computing power due to the project cycle. It is flexibly used on demand, allowing researchers to focus on scientific research and improve efficiency without being distracted by computing power and platform usage.
A higher institution of the first-class universities and disciplines of the world - large language model training
Providing large computing power base to ensure research efficiency
Cooperation background: The large model research group of a higher institution of the first-class universities and disciplines of the world needs to use more than 100 high-end GPU accelerator cards to train large language models. However, due to supply constraints, it is difficult for the research team to obtain sufficient high-end GPU computing power suitable for large model training in the short term. In order to address this problem, PARATERA launched an in-depth cooperation with the university, providing shared computing resources with 256 A100 GPUs to support the training needs of the large language model.
Cooperation benefits: Through this cooperation, the research team has avoided the situation of being unable to conduct scientific research due to insufficient computing power. PARATERA Smart Computing Cloud Team assisted the research team in setting up the computing environment and completing the migration and deployment of the model. In addition, it also provided 7×24 hours of technical support services to solve the needs of software dependencies, performance optimization and model debugging, so that researchers could focus on scientific research and improve research efficiency.
An AI large model enterprise
Improving computational efficiency based on super-computing architecture
Cooperation background: The main research scenario of this enterprise is the large model training of natural language processing, which not only requires ensuring the stable operation of the program, but also continuously optimizing the program performance during the operation process. The entire process requires a large amount of resources and computing power based on super-computing architecture to improve computational efficiency.
Cooperation benefits: By using the large model computing resources based on super-computing architecture, the customer's computing power requirements of natural language processing scenarios were satisfied. PARATERA Smart Computing Cloud not only helped the user optimize program performance and install environment, but also helped the user avoid computing interruption, video memory overflow, memory overflow, running error and other conditions during computing. While operating stably, it improved the computational performance of the 180 B scale model by 30%, improving computational efficiency.
A novel AI R&D institution
Trillion-level pre-training model, computing power service guarantee
Cooperation benefits: Based on PARATERA's computing power network services, A100 computing power resources based on "super-computing architecture" were quickly provided in a short period of time, and in addition, basic environment construction, model deployment, application operation characteristic performance analysis, and 7×24-hour services were provided for technical support guarantee and other services. Through the combination of "computing power + large model + technical services", the company created an efficient and stable intelligent computing cloud computing power base for the customer.
An innovative drug R&D enterprise
Accelerating R&D and reducing the project time by 50%
Cooperation background: An innovative drug R&D company, due to the rapid growth of its business, needed to use AI cutting-edge algorithms, combined with the experience of computational chemistry and medicinal chemistry, to quickly screen billions of molecules, with short R&D cycles, large calculations, high throughput, large storage capacity and other characteristics. Due to insufficient local GPU resources, the speed of computing resource expansion was difficult to match the growth needs of the project.
Cooperation benefits: The customer built a drug R&D platform on the PARATERA Smart Computing Cloud, which can not only support multi-team and multi-task parallel model training, but also schedule hundreds of NVIDIA V100 GPU resources in parallel for large-scale distributed training of Pytorch and Tensorflow, as well as molecular dynamics simulations such as GROMACS and LAMMPS. The project time was shortened by more than 50%, and the R&D cost was greatly reduced.