The AI laboratory proposes a standardized approach to the efficient use of infrastructure and software stacks for the management of projects that use complex neural networks.
Workloads will be delivered with flexibility in resource allocation, making safe accessible environments available to perform tasks characterized by:
The main server supporting the services offered is an NVIDIA DGX A100 server.
This solution offers a universal computing environment concerning the type of workload.
Resources to manage all workloads: training, inference, analytics;
Optimal use of resources (with very high peak performance 5 petaFLOPS);
Scalability by design. The system allows you to dynamically acquire only the resources you need by factoring in the resources;
Its architecture features 8x NVIDIA A100 GPUs with 320GB of GPU memory – 30TB Gen4 NVME SSD
In addition, the DGX A100 server supports Multi-Instance Virtual GPUs.
Each user can decide how many virtual GPUs to acquire and how much memory they need (slices) in a flexible and granular way.