HPE shares more details about the new Cray supercomputers and the new components they contain.
Hewlett Packard Enterprise (HPE) introduces new components for its Cray supercomputers. These include three new blades, new management software, and a faster networking solution. These innovations are designed to make the systems more powerful and easier to scale, especially for AI applications and heavy computational processes.
New Blades
The three new blades are designed for different types of tasks and combine computing power from NVIDIA and AMD GPUs and CPUs. All modules are liquid-cooled, allowing them to deliver high computing capacity in a small footprint. They can all be combined within the same Cray GX5000 rack.
- The GX440n Accelerated Blade uses NVIDIA processors and GPUs and has up to 192 GPUs per rack.
- The GX350a Accelerated Blade uses AMD processors and GPUs and has up to 112 GPUs per rack.
- The GX250 Compute Blade is designed for tasks without GPUs but with eight AMD processors, with up to 40 modules per rack.
HPE also introduces Slingshot 400, which is a new network component that supports 64 ultra-fast connections. According to HPE, this allows for the construction of large supercomputer networks that can exchange data quickly.
With AI in Mind
HPE also introduces new software to make supercomputer management easier. This software supports environments where multiple users or teams work simultaneously and offers features for installing, monitoring, and energy-efficient management of large systems. The software also works well with virtual machines and containers.
For storage, HPE launches the Cray Supercomputing Storage Systems K3000, a specialized system for AI applications. It uses the fast, open-source storage technology Distributed Asynchronous Object Storage (DAOS) and is available in variants designed for maximum speed or high capacity. The system can handle various types of networks.