High-Performance Computing Cluster (HCC) uses high-performance CVM instances as nodes, and instances are interconnected via Remote Direct Memory Access (RDMA). It offers high bandwidth and ultra-low latency network services, significantly improving network performance. It can meet the parallel computing needs of large-scale high-performance computing, AI, big data recommendations, and other applications.

Based on the elastic high-performance CVM architecture, it can achieve computation without virtualization loss and implement server features without loss.

Support fully automated production and auto scaling, featuring flexibility and high performance.

General CPU computing power supports online hybrid deployment of multiple architectures such as x86 and ARM, while intelligent computing GPU computing power supports NV and multiple localized GPUs, meeting the elastic usage requirements of different customers with different computing power.

Support flexible storage capacity expansion with COS or CFS and ensure strong consistency among the three replicas.

Provides ultra-fast I/O instances for selection, which cal be used together with local NVMe SSD, to meet the needs of I/O-intensive and on-premises caching scenarios.

Supports interconnection between nodes through Virtual Private Cloud (VPC) and non-blocking RoCEv2 RDMA networks, with transmission latency as low as 2 us, which can widely support Ethernet-based applications.

Use intelligent NICs and a VPC network to help users build cloud-native high-performance computing applications with HCC, CFS, and other services.

Comparison with Local Supercomputing Clusters

HCC has the following advantages over local supercomputing clusters:

Large-scale AI training, material simulation, industrial simulation (CAE), and other high-performance computing scenarios have very high requirements for the computational performance, stability, and real-time performance of servers.

High-performance CVM provides the latest generation of CPU and GPU instances, supporting GPU heterogeneous hardware acceleration for HPC job computing, with higher cost-effectiveness.

High-performance CVM computing cluster supports RDMA low-latency network interconnection based on RoCEv2, with latency as low as 2-3 us. It supports dynamic computing power scale-out across multiple servers, while achieving a near-linear computing speedup ratio.

Scenarios, such as autonomous driving, natural language processing (NLP), and recommendation systems, feature large data volume and intensive computing workload. HCC instances can support high-speed, low-latency RDMA network interconnection and intensive-workload computing by using the latest generation of CPU architectures and heterogeneous GPU components. This helps meet the business's needs for high computing performance, high stability, and real-timeliness of servers.

Many manufacturing enterprises in the automobile, aviation, and other industries need to use simulated computing to drive design. The high-performance computing clusters built by enterprises themselves require large investments and long cycles, and it is difficult to continuously meet the demand. HCC can be instantly deployed and automatically scaled. Through the high-speed, low-latency RDMA network and the latest CPU architecture, HCC implements parallel processing to quickly meet the simulation demands of enterprises in aerospace, industrial manufacturing, and other industries and promote product R&D promptly.

HCC can perform large-scale molecular dynamics simulations through high-speed, low-latency RDMA networks, predict and analyze the interactions and changes between biological protein molecules and lipid molecules, assisting in new drug research and development.

HCC can provide supercomputing platforms for universities and research institutions for numerical simulation, numerical computation, simulation verification, and other applications.

tencent cloud