This topic describes the features of Super Computing Cluster (SCC) instance families and lists the instance types of each family.

Introduction

SCCs are based on ECS Bare Metal Instances. With the high-speed interconnects of Remote Direct Memory Access (RDMA) technology, SCCs greatly improve network performance and the acceleration ratio of large-scale clusters. Therefore, SCCs have all the benefits of ECS Bare Metal Instances and can provide high-quality network performance that features high bandwidth and low latency.

SCCs are used to meet the demands of applications such as high performance computing, artificial intelligence, machine learning, scientific and engineering computing, data analysis, and audio and video processing. In the clusters, nodes are connected over RDMA networks that feature high bandwidth and low latency. This ensures the parallel efficiency of applications in areas such as high performance computing, artificial intelligence, and machine learning. The RDMA over Converged Ethernet (RoCE) rivals an InfiniBand network in terms of connection speed and can support more Ethernet-based applications.

The combination of SCCs and other Alibaba Cloud computing services such as ECS and Elastic GPU Service provides E-HPC with the ultimate high performance parallel computing resources, making supercomputing on the cloud possible.

Comparison of SCCs, physical machines, and virtual machines

The following table compares the features of SCCs, physical machines, and virtual machines. In this table, Y means supported, N means not supported, and N/A means not applicable.

Feature type Feature SCC Physical machine Virtual machine
Automated O&M Delivery within minutes Y N Y
Compute Zero performance loss Y Y N
Zero feature loss Y Y N
Zero resource preemption Y Y N
Storage Compatible with ECS disks Y N Y
Startup from system disks Y N Y
Quick reset of system disks Y N Y
Use of ECS images Y N Y
Cold migration between physical and virtual machines Y N Y
No need to install the operating system Y N Y
No need of local Redundant Arrays of Independent Disks (RAID), and better protection of data in disks Y N Y
Network Compatible with ECS VPCs Y N Y
Compatible with the ECS classic network Y N Y
No communication bottlenecks between physical and virtual machine clusters in VPCs Y N Y
Management Compatible with existing ECS management systems Y N Y
Consistent user experience on features such as VNC with that on virtual machines Y N Y
Out-of-band (OOB) network security Y N N/A

scch5, SCC instance family with high clock speed

Features
  • I/O optimized.
  • Supports only standard SSDs and ultra disks.
  • Supports both RoCE and VPCs, of which RoCE is dedicated to RDMA communication.
  • Provides all the features of ECS Bare Metal Instances.
  • Equipped with 3.1 GHz Intel ® Xeon ® Gold 6149 (Skylake) processors.
  • Offers a CPU-to-memory ratio of 1:3.
  • Scenarios:
    • Large-scale machine learning training
    • Large-scale high performance scientific computing and simulations
    • Large-scale data analysis, batch processing, and video encoding
Instance types
Instance type vCPU Physical cores Memory (GiB) GPU Bandwidth (Gbit/s) Packet forwarding rate (Kpps) RoCE (Gbit/s) IPv6 support NIC queues ENIs (including one primary ENI) Private IP addresses per ENI
ecs.scch5.16xlarge 64 32 192.0 None 10.0 4,500 2 × 25 No 8 32 10
Note

sccg5, general purpose SCC instance family

Features
  • I/O optimized.
  • Supports only standard SSDs and ultra disks.
  • Supports both RoCE and VPCs, of which RoCE is dedicated to RDMA communication.
  • Provides all the features of ECS Bare Metal Instances.
  • Equipped with 2.5 GHz Intel ® Xeon ® Platinum 8163 (Skylake) processors.
  • Offers a CPU-to-memory ratio of 1:4.
  • Scenarios:
    • Large-scale machine learning training
    • Large-scale high performance scientific computing and simulations
    • Large-scale data analysis, batch processing, and video encoding
Instance types
Instance type vCPU Physical cores Memory (GiB) GPU Bandwidth (Gbit/s) Packet forwarding rate (Kpps) RoCE (Gbit/s) IPv6 support NIC queues ENIs (including one primary ENI) Private IP addresses per ENI
ecs.sccg5.24xlarge 96 48 384.0 None 10.0 4,500 2 × 25 No 8 32 10
Note

sccgn6, compute optimized SCC instance family with GPU capabilities

Features
  • I/O optimized.
  • Offers a CPU-to-memory ratio of 1:4.
  • Equipped with 2.5 GHz Intel ® Xeon ® Platinum 8163 (Skylake) processors.
  • Provides all the features of ECS Bare Metal Instances.
  • Storage:
    • Supports enhanced SSDs (ESSDs), standard SSDs, and ultra disks
    • Supports a high performance CPFS
  • Networking:
    • Supports VPCs
    • Supports the RoCE v2 network, which is dedicated to low-latency RDMA communication
  • Uses NVIDIA V100 GPU processors that have the SXM2 module:
    • Powered by the new NVIDIA Volta architecture
    • Offers a 16 GB HBM2 GPU memory.
    • CUDA Cores 5120
    • Tensor Cores 640
    • Offers a GPU memory bandwidth of up to 900 GB/s
    • Supports up to six NVLink connections and total bandwidth of 300 GB/s (25 GB/s per connection)
  • Scenarios:
    • Ultra-large-scale machine learning training on a distributed GPU cluster
    • Large-scale high performance scientific computing and simulations
    • Large-scale data analysis, batch processing, and video encoding
Instance types
Instance type vCPU Memory (GiB) Local storage (GiB) GPU Bandwidth (Gbit/s) Packet forwarding rate (Kpps) RoCE (Gbit/s) IPv6 support NIC queues ENIs (including one primary ENI) Private IP addresses per ENI
ecs.sccgn6.24xlarge 96 384.0 None 8 × V100 30 4,500 2 × 25 Yes 8 32 10
Note

Billing method

SCCs support pay-as-you-go and subscription billing methods. For more information, see Comparison of billing methods.