The allocation of compute resources among various teams/projects is a common struggle among companies utilizing on-premise infrastructure. Efficiently allocating these resources can have System Administrators working overtime. Bursting workloads to the cloud to compensate for resource inflexibility has become common practice. On-Demand instances are especially prevalent for workloads that require GPU. Luckily, one of GVM Server’s main features helps solve these issues.
Organization-level provisioning is a nested roles feature that allows organizations to manage data and resources hierarchically for teams/projects. System Administrators can assign compute resources to Users, who can then utilize their resources as needed in virtual machines.
Admins can create and delete user accounts and manage all available compute resources (GPU, CPU, RAM, Storage). Admins can allocate specific resources to individual users, who can then utilize their resources in configurable virtual machines. Super Admin accounts will also be available to establish a more intricate organizational hierarchy.
User accounts are simple enough to be used by anybody, no matter their technical abilities. Once assigned resources by an admin, all users have to do is sign in through any browser and access their virtual machine. Users can also view the compute resources allocated to them and spin up virtual machines personalized for their workloads.
Company A is an AI startup that's working on multiple projects involving the training of complex neural network models. The company’s on-premise infrastructure consists of a single Supermicro server that contains 9 Intel Data Center Flex Series GPUs. This server came with Arc Compute’s GVM Server preinstalled.
The SysAdmin, using their Super Admin account, signs into GVM Server and creates Admin accounts for all three projects, and allocates 4 GPUs to Project 1, 2 GPUs to Project 2, and 3 vGPUs to Project 3, along with the other resources required (vCPUs, RAM, Memory, etc.).
The Project 1 Manager signs into their Admin account and creates one User account. That User is assigned a single virtual machine and allocated four vGPUs (4 x 40GB VRAM). The data scientist (User) working on Project 1 signs into the User account and starts training their neural network model in the virtual machine.
The Project 2 Manager signs into their Admin account and creates two User accounts, both assigned virtual machines allocated 1vGPU (40GB VRAM). The data scientists working on Project 2 sign in to their User accounts and start training their neural network models in the virtual machines.
The Project 3 Manager signs into their Admin account and creates four User accounts, all are assigned virtual machines allocated 1vGPU (two VMs allocated 40GB of VRAM and two allocated 20GB of VRAM). The data scientists working on Project 3 sign in to their User accounts and start training their neural network models in the virtual machines.
Thanks to GVM Server, Company A can take complete control of its compute resources. GVM Server virtualizes both CPUs and GPUs, enabling much higher resource utilization. Better resource utilization helps reduce Company A’s need to burst workloads to the cloud, resulting in considerable cost savings over time.
Do you think GVM Server could help your company manage its on-premise infrastructure? Check out the GVM Server - Solution Brief below + contact me with any questions/inquiries at erik@arccompute.io.
.jpg)
