Dardel Expanded with More Disk Space and NVIDIA GPUs

The Dardel system is now finally fully accepted! Due to various delays with deliveries during the installation period of the whole system (which spanned several years), PDC has received substantial compensation over that time. The latest compensation for delays, along with some unused funding from the initial investment, will now be used to expand Dardel with additional fast Lustre storage and some nodes with NVIDIA Grace Hopper GPUs.

The Lustre file system will be expanded with an additional 4.7 PB of hard disk storage and 260 TB flash storage. This will give a total Lustre capacity of 23.6 PB.

Many researchers who use Dardel have asked for NVIDIA GPU nodes to be made available at PDC. This is because many research applications are only available for the NVIDIA software stack with CUDA. In response to these requests, eight nodes will be added to Dardel, each with four Hopper (H100) GPUs and four NVIDIA Grace ARM CPUs. This will add a total of 32 NVIDIA Grace Hopper Superchips. Each of these nodes will have four Slingshot ports and the total expansion is expected to increase the performance of the system by around 1.3 PFLOPS, as measured with the high-performance Linpack (HPL) benchmark. For more information about the Grace Hopper Superchips, see www.nvidia.com/en-gb/data-center/grace-hopper-superchip .

Above: The photo shows an HPE Grace Hopper Superchip board with two nodes. Each of the nodes contains four NVIDIA Grace Hopper GPUs. The board is made by HPE using the NVIDIA Grace Hopper Superchips.