8VLink AI Computing Platform is committed to lowering enterprises’ barriers to accessing AI computing power, enabling an efficient, cost-effective, standardized and traceable AI application development process. It helps enterprises rapidly leverage network resources, GPUs, storage systems, machine learning frameworks and models supported by intelligent computing centers.
Diversified computing hardware and forms to meet application demands of various scenarios
Compatible with multiple frameworks and software stacks to satisfy diverse model training requirements
One-stop full-process coverage of data management, training and inference
Graphical low-code and zero-code workflow orchestration tools
Professional tool library reduces model R&D costs and improves development efficiency
Efficient large-scale model training with thousands or tens of thousands of GPUs
|
Resource IntegrationIntegrate multiple types of resources such as GPU, CPU, storage and network to build a high-speed parallel storage system and high-speed InfiniBand network, forming a heterogeneous GPU computing resource pool. The resource pool supports flexible cross-region allocation, and builds an efficient, low-latency dedicated AI network environment. It adapts to model parameter storage requirements, ensures efficient access of massive data and multi-server communication, greatly improves data transmission speed, and guarantees safe and stable data transmission as well as task execution. |
|
Distributed SchedulingBased on container technology, it realizes efficient scheduling of various resource types and automatically allocates and manages GPU computing resources, improving the efficiency of resource and task scheduling. The system also provides resource group and priority configuration functions, effectively shortening the scheduling path of data transmission to meet the training and inference demands of large language models. In addition, the platform supports continuous operation of model fine-tuning and inference services. |
|
Heterogeneous SupportRealize unified management of multiple heterogeneous computing resources, integrate mainstream domestic and overseas resources such as GPU, TPU and DPU, and create a centralized computing power resource library. According to the demands of different computing tasks, the system can flexibly schedule and allocate computing resources, providing multiple deployment options including computing resource groups, whole-server rental and card-level application. |
|
Training and Inference IntegrationProvide comprehensive AI services including data labeling, dataset management, algorithm construction, model training, model optimization, model management and deployment inference. The platform is built-in with a variety of commonly used GPU function libraries and toolkits, supporting mainstream training frameworks such as TensorFlow, PyTorch and PaddlePaddle. |
|
AI RepositoryThe platform supports image repository, algorithm repository and data sample repository to store data and codes required for user training and inference. |
Enable flexible access and management of various computing resources, with dynamic allocation and scheduling based on leasing demands
Support large-scale scientific computing and simulation experiments, providing researchers with abundant computing resources
Support multiple financial analysis models and algorithms to deliver accurate risk assessment, market forecasting and investment decision support
Compatible with various simulation models and scenario configurations, offering users rich simulation experiment resources and environments
浙公网安备33010802008444号
