The Nectar GPU service has been made available for researchers to access high end resources which would not otherwise be available to them. All projects with a national allocation status are eligible to use this service. You may also be eligible via your local node/institution. Contact our helpdesk for more information.
For more information about what a GPU is, view this video. Additionally, this external beta Carpentries course which includes an explanation about GPU's, and GPU programming tutorials.
The GPUs are available to use via making a booking using the Nectar Reservation Service.
Currently 2 flavor classes are provided:
The specifications (RAM, VCPUs etc), are available on our Nectar flavors page.
Note: All flavors use Nvidia virtual GPU technology meaning the GPU your instance has access to is shared. All flavors have a minimum guaranteed slice alongside dedicated GPU RAM with bigger flavors having a larger minimum slice of computing power.
Using a GPU instance
Once you have made a reservation, you need to run through the standard process to launch an instance, by selecting the appropriate flavor and image.
The flavor which contains the particular GPU resources you reserved can be found on the detail page for the reservation. You can find this by clicking the reservation ID.
It would look something similar to what’s pictured below.
When you are in the launching instance dialog, you should see it appear on the flavor tab.
The image, which is usually the operating system you add to an instance, is a specially built one for use with a GPU. You need to select the one called NeCTAR Ubuntu 20.04 LTS (Focal) amd64 (NVIDIA vGPU). As per its name, it is our standard Nectar Ubuntu 20.04 image, but includes the NVIDIA vGPU driver installed.
Similarly, in the launching instance dialog, on the source tab, you should see it appear in the list of images.
The provided image also contains the appropriate licensing required to run the GPU hardware, which is automatically configured for you. If the current image isn’t sufficient, or you want to use your own image, reach out to us for help. You can lodge a support ticket.
Once you have launched your instance, you can check the GPU and provided license is active by using the following command:
The output will contain information about the attached GPU, and at the bottom, should show the licensed status as `licensed`.