Powerful GPU Cluster for AI computing

Make the most of the HPC Virtual Server service. You get access to a virtual server with pre-installed Ubuntu or AlmaLinux operating system and a high-performance NVIDIA HGX A100 graphics card.

The NVIDIA A100 graphics card has 8 GPUs (graphics cores), with 1GPU having a memory size of 80 GB and performance for "HPC and AI compute FP64" = 19.5 TF (TF = teraFLOPS). This powerful graphics card will enable even the most demanding computing tasks to be handled with ease and speed.

The service is offered in virtual server templates according to the number of GPUs which the customer orders. The provider offers the possibility to use the service guaranteed or unguaranteed. These options differ in the amount of payment for the provided computing resources.

The service is suitable for companies that use high computing power, especially for the purposes of neural network learning, analysis of large datasets, scientific computing or image and video processing.
 

For guaranteed GPU computing power, the customer chooses between a monthly virtual server rental or an hourly rental, regardless of whether or not they perform computing tasks using the GPU. At all times, the provider guarantees the customer the allocated number of GPUs. The customer pays the full rate for the computing power.

For non-guaranteed GPU computing power, the customer only uses an hourly rental. If the customer does not use the GPU for computing tasks, the provider can utilise the GPU for another customer by removing the GPU from the virtual server. When the GPU is released by another customer, it is returned to the original virtual server by agreement. The customer pays a reduced hourly rate for the time that the GPU is not used for computing power.

The customer receives the following accesses when setting up the service:

• SSL VPN for users,
• vCenter access to the assigned virtual server,
• SSH access to the virtual server running the operating system.

Bespoke software add-ons:

Due to the wide possibilities of using GPU computing power for AI/ML/NN, the customer can arrange with the provider to install the following software tools.

For GPU calculations:

• Apache Spark 3.0 + (supports GPU)
• TensorFlow
• PyTorch
• Knime/Knime Server
• NVIDIA Triton Server

For administration:

• Label Studio
• MLFlow
• GitLab

For work with data:

• NAS: NFS/CIFS (TrueNAS)
• Object storage: S3 (MinIO, TrueNAS)
• Block storage: iSCSI (TrueNAS)
• DB: PostgreSQL, CitusData

Basic components of the service:

• virtual server template with GPU
• static public IP address
• SSL VPN accesses
• SSH accesses to the virtual server operating system
• vCenter accesses to the assigned virtual server
• installation according to customer specifications

Additional components of the service:

• SATA 10k disk space expansion
• SSD disk space expansion
• Green Drive private data storage
• installation of container platform according to customer specification
• specialist consulting work

The service is charged for the ordered virtual server computing templates on a monthly basis according to the customer's choice in the form of a full monthly rental or per consumed hours. In addition, for installation and consulting work, public IP address and according to the ordered additional service components.