Create a cloud server with GPU
GPUs (graphics processing units) can be added to a cloud server — when creating a server or to an existing server.
GPUs are used as dedicated PCI devices inside the cloud server.
GPUs are available:
- in fixed-configurations of the GPU Line;
- and arbitrary configurations.
GPU configurations can only be used with the server's network boot disk.
Create a server with GPU
Use the Create cloud server instruction.
Select:
- source is a finished image of
Ubuntu 22.04 LTS Machine Learning 64-bit. The image contains the drivers needed to work with graphics processors. If you choose another source, you will need to install the drivers yourself; - configuration — with vCPUs from 2 cores: fixed GPU Line configuration or arbitrary.
Add a GPU to an existing cloud server
If the cloud server has an arbitrary configuration and a network boot disk, GPUs can be added to it.
- In Control Panel, go to Cloud Platform → Servers.
- Open the server page → Configuration tab.
- In the Configuration box, select an arbitrary configuration.
- Click Add GPU.
- Specify the number of GPUs.
- Click Save and Reload.
Available GPUs
You can view the current list of GPUs in control panel: under Cloud Platform → Servers → click Create Server.
You can see the availability of GPUs in regions in the [GPUs for cloud servers] availability matrix(/control-panel-actions/availability-matrix.mdx#gpu-for-cloud-and-kubernetes).
NVIDIA® A100
Has maximum performance for AI, HPC, and data processing. Suitable for deep learning, scientific research and data analytics.
Ampere® based, up to 2 TB/s throughput. Refer to NVIDIA® documentation for detailed specifications.
Fixed GPU Line configurations are available from 1 to 8 GPUs × 40 GB, with vCPUs from 6 to 48, RAM from 87 to 700 GB.
Random configurations are 1 to 8 GPUs × 40 GB, with vCPUs from 2 to 32, RAM from 512 MB to 256 GB.
NVIDIA® A100 NVLink
You can combine two NVIDIA® A100 GPUs using NVLink technology.
NVLink accelerates the communication speed of GPU interconnect compared to the PCIe interface. NVLink-coupled GPUs allow more memory to be utilized and improve server performance for complex computations, such as training large language ML models.
NVLink works with NVIDIA® A100 — Ampere®-based GPUs with up to 2 TB/s of bandwidth. For detailed specifications of the NVIDIA® A100 and a description of NVLink technology, refer to the NVIDIA® documentation.
In fixed GPU Line configurations, 2 GPUs × 40 GB are available, with 24 vCPUs, 128 GB RAM. If configurations with more resources are needed, create a ticket.
NVIDIA® Tesla T4
Suitable for Machine Learning and Deep Learning, inference, graphics work and video rendering. Works with most AI frameworks and is compatible with all types of neural networks.
Based on Turing®, up to 300 GB/s throughput. Refer to NVIDIA® documentation for detailed specifications.
In fixed GPU Line configurations, 1 to 4 GPUs × 16 GB are available, with vCPUs from 4 to 24, RAM from 32 to 320 GB.
Random configurations are 1 to 4 GPUs × 16 GB, with vCPUs from 2 to 32, RAM from 512 MB to 256 GB.
NVIDIA® A30
Suitable for AI-inference, HPC, language processing, conversational artificial intelligence, recommender systems.
Ampere® based, up to 933 GB/s bandwidth. Refer to NVIDIA® documentation for detailed specifications.
Fixed GPU Line configurations are available from 1 to 2 GPUs, with vCPUs from 16 to 48, RAM from 64 to 320 GB.
In random configurations, 1 to 2 GPUs, with vCPUs from 2 to 32, RAM from 512MB to 256GB.
NVIDIA® A2
Entry-level GPU. Suitable for simple inference, video and graphics, Edge AI (edge computing), Edge video, mobile cloud gamification.
Ampere® based, up to 200 GB/s bandwidth. Refer to NVIDIA® documentation for detailed specifications.
In fixed GPU Line configurations, 1 to 4 GPUs × 16 GB are available, with vCPUs from 12 to 48, RAM from 32 to 320 GB.
Random configurations are 1 to 4 GPUs × 16 GB, with vCPUs from 2 to 32, RAM from 512 MB to 256 GB.
NVIDIA® GTX 1080
Performance and power efficient GPU. The solution is implemented with FinFET technology and GDDR5X memory. Dynamic load balancing helps to separate tasks so that resources don't sit idle waiting. Featuring maximum performance for information display, VR, ultra-high resolution settings and data processing.
Pascal® based, up to 320.3 Gb/s throughput. Refer to NVIDIA® documentation for detailed specifications.
In fixed GPU Line configurations, 1 to 8 GPUs × 8 GB are available, with vCPUs from 8 to 28, RAM from 24 to 96 GB.
Random configurations are 1 to 8 GPUs × 8 GB, with vCPUs from 2 to 32, RAM from 512 MB to 256 GB.
NVIDIA® A2000
Power-efficient GPU for compact workstations. Suitable for AI, graphics and video rendering.
Ampere® based, up to 288 GB/s bandwidth. Refer to NVIDIA® documentation for detailed specifications.
In fixed GPU Line configurations, 1 to 4 GPUs × 6 GB are available, with vCPUs from 6 to 24, RAM from 16 to 320 GB.
Random configurations are 1 to 4 GPUs × 6 GB, with vCPUs from 2 to 32, RAM from 512 MB to 256 GB.
NVIDIA® A5000
A versatile GPU, suitable for any task within its performance limits.
Ampere® based, up to 768 GB/s bandwidth. Refer to NVIDIA® documentation for detailed specifications.
In fixed GPU Line configurations, 1 to 2 GPUs × 24 GB are available, with vCPUs from 8 to 48, RAM from 32 to 320 GB.
Random configurations are 1 to 2 GPUs × 24 GB, with vCPUs from 2 to 32, RAM from 512 MB to 256 GB.