What is Al-Farabium
Al-Farabium is a GPU cloud based on a supercomputer with NVIDIA H200 graphics processors deployed in data centers across Kazakhstan. You get access to computing resources for running and training neural networks with hourly billing, no limits on workload size, and no need to deploy your own infrastructure.
Cluster specifications
- 400 NVIDIA H200 GPUs
- 100 TB of RAM
- NVIDIA reference architecture
- 56 TB of GPU memory
- 26.8 petaflops (FP64) for AI model training
- 6 PB of high-performance storage
- 1,583 petaflops (FP8) for AI inference
- InfiniBand network with 665 TB/s throughput
Server configurations
Virtual GPU server
• 1 NVIDIA H200 GPU (virtualized)
• 142 GB of GPU memory
• 48 vCPU cores (3.25 GHz)
• 256 GB of DDR5 RAM
• 3 TB of NVMe SSD storage
• 7 TB of HDD storage
• 142 GB of GPU memory
• 48 vCPU cores (3.25 GHz)
• 256 GB of DDR5 RAM
• 3 TB of NVMe SSD storage
• 7 TB of HDD storage
Bare-metal server
• 8 NVIDIA H200 GPUs
• 1.128 TB of GPU memory
• 384 vCPU cores
• 2 TB of RAM
• 56 TB of HDD storage
• 1.128 TB of GPU memory
• 384 vCPU cores
• 2 TB of RAM
• 56 TB of HDD storage
Additional options
Available on request: • NVMe SSD storage
• HDD storage
• Support for a block of 8 static IP addresses
• HDD storage
• Support for a block of 8 static IP addresses
Project goals and objectives
The strategic goal is to establish a sovereign artificial intelligence infrastructure in Kazakhstan that is accessible to developers, researchers, and government organizations.
Ensure that data is stored and processed within the country
Provide developers and organizations with direct access to high-performance computing resources
Support the development of a regional center for artificial intelligence and big data
Reduce dependence on foreign cloud providers and services
Accelerate the adoption of AI technologies in the public sector, education, healthcare, and business
Ensure compliance with national requirements for data protection and data localization
User capabilities
Compute capacity rental based on NVIDIA H200 GPUs
Deployment, training, and scaling of models with no limitations
Connectivity via secure communication channels (VPN, L2/L3)
Creation of a virtual IT infrastructure (VDC) tailored to specific workloads
In-country data processing with support for hybrid and multicloud scenarios
Integration with existing digital services and platforms
Use cases
Education
AI course delivery, student training, integration with I–Mektep and the National Education Database
Healthcare
Medical image analysis, telemedicine, integration with MED365
Public sector
Video surveillance, chatbots, digital twins, analytics for Smart City initiatives
Business
Content generation, recommendation systems, machine learning
Telecom and IT services
Integration with AituCloud, BTS Digital, and the TV+ platform
Innovation
Unmanned systems, video tokens, speech technologies, and ML prototyping
