The pass-through GPU performance provides excellent parallel computing capacities. The GPU Virtual Machines of different instance types can be selected according to business scales and the instance specifications can be adjusted whenever depending on the business scale.
At most 4 pieces of Nvidia® Tesla® V100 GPU can be matched. The peak value of single-precision floating-point arithmetic per machine is up to 64 TFlops, while the peak value of double-precision floating-point arithmetic is up to 28TFlops. The mixed-precision calculation capacity up to 112TFlops is provided. Totally, the 64GB HBM2 video memory is possessed, and the Intel® Xeon® E5 2650v4 process and the DDR4 memory are used in combination.
At most 4 pieces of Nvidia® Tesla® P40 GPU can be matched. The peak value of single-precision floating-point arithmetic per machine is up to 48 TFlops; the int8 integer arithmetic peak value can reach up to 188Tops; totally, the 96GB GDDR5 video memory is possessed; and the Intel® Xeon® E5 2683v4 process and the DDR4 memory are used in combination.
The local temporary data disk is configured, which can be the local HDD disk or the local SSD disk, to provide extremely-low access time delay and excellent IO performance. This data disk is applicable for saving cache and other temporary data and combining general SSD Disk, performance type SSD Disk or capacity-oriented HDD Cloud Disk. When providing high-availability data storage architecture, the best data access performance can be exerted.
The GPU Virtual Machines can provide you with maximum 10-gigabit network environment, guarantee the application response speed with extremely-low time delay, and can meet demands for network performance under different scenarios. Support self-planning for network deployment on demand, including setting VPC with/without preset network range, subnet segment, routing policy, etc., and realize multi-level security protection via security groups and network ACL.
Manage the GPU Virtual Machines via two means (OpenAPI and the console) by creating, starting, stopping and other ways. At the same time, two methods are supported to log in the GPU Virtual Machines for environmental configuration, i.e. the console VNC and remote SSH.
The GPU Virtual Machines supports two billing methods, i.e. the Monthly Package and the Pay By Configuration. The Monthly Package is a Pay-In-Advance mode that the expense of 1-10 months or 1-3 years is paid at one time. The Monthly Package is suitable for the scenario when the business demand can be estimated in advance (suitable for scenarios when the instance quantity and scale can be estimated in advance). The price of the Monthly Package is more favorable than that of the Pay By Configuration mode. However, the Pay By Configuration is the Pay-As-You-Go mode, providing more flexible life circle management. The billing period of such mode is 1 hour, and the settlement is made based on the Virtual Machine specification and use duration (accurate to seconds). Switch from Pay By Configuration to Monthly Package is supported.