By Glenn Satchell
ARDC recently provided funding for the expansion of our Nectar Node with new GPU and large memory compute servers! This is in addition to the hardware refresh completed in 2021.
The new GPU servers each include either two NVIDIA A100 GPUs or two A40 GPUs. The A100 Tensor Core GPU with 80GB of memory is perfect for Machine Learning. The A40 with 48GB of memory is great for inference, ray tracing, graphics and simulation workflows. These servers also have NVME storage that can feed data to the GPUs at an astounding speed to keep the pipeline running at maximum speed.
The new huge memory servers contain 4 Terabytes of RAM and are suitable for virtual machines with 1TB or more of memory.
The A100 can be split into several virtual GPU configurations with either 4, 2 or 1 GPU cores. With two cards this allows up to 14 virtual machines to have a GPU. Larger portions of the GPU are also possible.
The A40 can be split into 6 virtual GPUs allowing for accelerated desktops or enhanced compute for graphical and simulation workflows.
As these are scarce resources, access will be managed through a scheduling service provided by the Openstack Blazar project. This will allow researchers to book a period of time to have access to the systems. The goal is to provide fair access for all users.