- Engineering & Technology
The Generative AI Computing Center (GenIA) is designed as a modular and flexible infrastructure for training, fine-tuning and deploying large-scale generative AI foundation models (text, images, video, audio, code, etc.).
Its architecture is based on a hypervisor that enables computing resources to be dynamically configured according to each use case, ensuring the availability, accessibility and security of data and models.
The memory capabilities of current GPUs are expanded through new nodes equipped with NVIDIA A100 and/or H100 GPUs (or equivalent), capable of efficiently handling very large models, with a focus on scalability and reducing downtime.
Equipment
- High-performance computing center based on servers with computing nodes equipped with NVIDIA A100 and/or H100 GPUs (or equivalent) for training and fine-tuning foundation models
- Total storage capacity of 130–150 TB (NVMe for the system and Ceph, SATA SSD for general-purpose use)
- CPU per node: 2 × Intel Xeon Gold 6430 32C/64T or higher
- 1 TB of RAM, expandable up to 2 TB per node
- Network connectivity with at least two 10G SFP+ links per server
- Hypervisor and virtualization software for dynamic resource allocation
- Prototyping, training and fine-tuning of generative AI models using private data
- Deployment of complete generative AI (GenAI) pipelines (data preparation, training, evaluation and operation) leveraging HPC infrastructure
- Secure experimentation with foundation models without requiring companies to own their own infrastructure
- Support for R&D projects in generative AI and advanced computing
- Advice on the responsible use of generative AI, including privacy, security and data governance
VICOMTECH
Gipuzkoa Science and Technology Park, Paseo Mikeletegi 57, 20009 Donostia / San Sebastián (Gipuzkoa)
Data and AI Platforms:
- AI, Big Data and Data Science
- Digital Technologies
Application sectors
- Automotive
- Energy
Industry plan
- IRABAZI · Automotive Components
- IRABAZI · Energy
