HPC Accelerated Compute Service
Overview
This service provides GPU-accelerated HPC nodes for large-scale AI training, fine-tuning, inference, and other compute-intensive workloads. It offers managed access to high-performance accelerator resources for users who need scalable compute capacity for modern machine learning and demanding numerical applications. The service is based on a managed accelerated compute environment with GPU nodes, high-speed interconnects, container-compatible execution options, batch scheduling, and operational monitoring, designed to support AI training workflows and tightly integrated HPC workloads.
Outcome
Efficient execution of large-scale AI training, accelerated simulations, and high-throughput compute workloads with access to managed GPU resources.
Service Card
| Category | Infrastructure and data-related services |
| Domain | GPU-accelerated HPC and AI Compute Infrastructure |
| Vertical | Cross-domain |
| Target users | AI Researchers, Data Scientists, HPC Users, ML Engineers, SMEs, and Startups |
| Service type | Infrastructure Service / Managed |
| Outputs | Managed access to GPU-enabled HPC queues and execution environments for training, inference, and accelerated compute jobs. |
| Requirements | Project registration, approved access credentials, and workload preparation for GPU-accelerated execution. |
| Limitations | GPU capacity is a shared resource and is scheduled to ensure fair usage, taking into account job size, execution time, and other relevant parameters. Non-accelerated workloads are only supported by the CPU compute service. The current GPU resources include A100 40 GB HBM2 PCIe 4.0 cards. For very large training models, adjustments may be required, and workloads with high inter-GPU communication may experience moderate performance differences compared to tightly coupled multi-GPU systems. |
| Availability | Available (new system Q3/2027) |
| Licensing | |
| Visual content | |
| Visual caption |