GPU virtualisation streamlines server deployment
NVIDIA’s virtual GPU (vGPU) now supports server virtualisation for AI, deep learning and data science. Previously limited to CPU-only, AI workloads can now be easily deployed on virtualised environments like VMware vSphere with new vComputeServer software and NVIDIA NGC.
Through a partnership with VMWare, this architecture will help organisations to seamlessly migrate AI workloads on GPUs between customer data centres and VMware Cloud on AWS.
vComputeServer gives data centre administrators the option to run AI workloads on GPU servers in virtualised environments for improved security, utilisation and manageability. IT administrators can use hypervisor virtualisation tools like VMware vSphere, including vCenter and vMotion, to manage all their data centre applications, including AI applications running on NVIDIA GPUs.
Many companies deploy GPUs in the data centre, but GPU-accelerated workloads such as AI training and inferencing run on bare metal. These GPU servers are often isolated, with the need to be managed separately. This limits utilisation and Through a partnership with VMWare, this architecture will help organisations to seamlessly migrate AI workloads on GPUs between customer data centres and VMware Cloud on AWS.
With vComputeServer, IT admins can better streamline management of GPU-accelerated virtualised servers while retaining existing workflows and lowering overall operational costs. Compared to CPU-only servers, vComputeServer with four NVIDIA V100 GPUs accelerates deep learning 50x faster, delivering performance near bare metal.
The release brings support to VMware vSphere along with existing support for KVM-based hypervisors including Red Hat and Nutanix. This allows admins to use the same management tools for their GPU clusters as they do for the rest of their data centre.
By expanding the vGPU portfolio with NVIDIA vComputeServer, NVIDIA is adding support for data analytics, machine learning, AI, deep learning, HPC and other server workloads. The vGPU portfolio also includes virtual desktop offerings — NVIDIA GRID Virtual PC, and GRID Virtual Apps for knowledge workers and Quadro Virtual Data Centre for professional graphics.
NVIDIA vComputerServer provides features like GPU sharing, so multiple virtual machines can be powered by a single GPU, and GPU aggregation, so one or multiple GPUs can power a virtual machine. This results in maximised utilisation and affordability.
Features of vComputeServer include:
- GPU Performance: Up to 50x faster deep learning training than CPU-only, similar performance to running GPU on bare metal.
- Advanced compute: Error correcting code and dynamic page retirement prevent against data corruption for high-accuracy workloads.
- Live migration: GPU-enabled virtual machines can be migrated with minimal disruption or downtime.
- Increased security: Enterprises can extend security benefits of server virtualisation to GPU clusters.
- Multi-tenant isolation: Workloads can be isolated to securely support multiple users on a single infrastructure.
- Management and Monitoring: Admins can use the same hypervisor virtualisation tools to manage GPU servers, with visibility at the host, virtual machine and app level.