We’re excited to present our ambitious new project: the Gcore Generative AI Cluster based on twenty servers powered by NVIDIA GPU. The cluster offers the significant performance boost required for training and inference of large AI/ML models, including those for generative AI. As a thriving center of technological innovation, Luxembourg provides the perfect setting for this monumental undertaking.
Powered by Leading GPU Accelerators
The Gcore Generative AI Cluster currently consists of twenty servers equipped with NVIDIA A100 Tensor Core GPUs. The A100 is a powerful accelerator that outperforms GPU competitors in most AI training tasks. It also delivers up to 100x faster AI training performance than modern CPUs.
Here is an A100-based server configuration:
- 8x NVIDIA A100 SXM 80 GB
- 2x Intel Xeon 8468
- 2 TB RAM
- 8×3.84 TB NVMe
- InfiniBand 800 Gbps
InfiniBand interfaces provide direct, high-speed GPU connections, ideal for scaling generative AI workloads.
Gcore GPU Infrastructure Types
You can choose how to use the cluster capacity as part of your preferred Gcore service. NVIDIA A100 GPUs power the following:
- Virtual Instances
- Bare Metal servers
- Managed Kubernetes based on Virtual Instances
- Managed Kubernetes based on Bare Metal
Gcore’s systems are optimized for the NVIDIA hardware to ensure peak performance.
Our setup integrates seamlessly with MLOps platforms like UbiOps, streamlining your machine learning workflows from experimentation to deployment. This allows you to manage all your models in one place, complete with version control, logging, auditing, and monitoring. You can track the usage and performance metrics of your models, setting needed alerts and notifications. Let your data work for you, not the other way around. For more details, see our announcement of the Gcore, Graphcore, and UbiOps partnership.
Gcore’s AI Plans for the Future
In the coming months, we’ll add 25 A100 and 128 H100 GPU-based servers to the Gcore Generative AI cluster. The H100 Tensor Core GPU is NVIDIA’s latest data center-class AI accelerator, delivering up to 4x faster AI training performance than the A100.
Here is an H100-based server configuration:
- NVIDIA HGX H100 8-GPU SXM GPUs Assembly
- 2x Intel Xeon Platinum 8480+
- 2 TB RAM
- 8x 3.84 TB Enterprise NVMe RI
- InfiniBand 3.6 Tbps
Stay tuned for our upcoming Global Intelligence Pipeline, designed to accelerate inference at the edge. With this project, we’re breaking down the barriers between data collection and real-time analytics, enabling smarter, faster decisions.
Get Started Now
To learn more about how GPUs drive generative AI, read our blog post.