Red Hat Virtualization: Supporting multiple NVIDIA virtual GPU workflows in virtualized environment
With two teenage sons who are passionate about video games, I can’t help but notice NVIDIA and its graphics processing units (GPUs). I’ve learned over the years that more powerful GPUs are continually sought after by gamers for better video game performance. But graphic rendering for video games is not the only application for GPUs: Professional visualization, high performance computing, big data, machine learning, and artificial intelligence are a few leading compute-intensive use cases for GPUs. Multiple GPUs are often deployed to speed interactive rendering of photorealistic images and to accelerate computational performance of real time simulations during product design, as well as to run ground-breaking research and applications.
GPU use cases have evolved from having a dedicated or integrated graphics card in a single physical workstation or server to completely virtual environments where a virtual machine manages a part of the physical GPU, called a virtual GPU (vGPU). The vGPU is used to perform tasks such as video capturing, encoding, graphic rendering and compute, providing a way to offload graphic rendering and compute-intensive tasks from CPUs within a virtual machine. This also helps to provide end users with improved access to performance-enhancing GPU features much in the same way as they do with physical systems.
If you take this concept a step further, just as it is possible to run multiple GPUs in a system, it should be possible to work with multiple vGPUs from within a virtual machine, right?
Well, that functionality has not existed until now. On October 9th, NVIDIA announced their virtual GPU (vGPU) October 2018 release (vGPU 7.0), the newest edition of their virtual GPU software that now supports multiple vGPUs in a single virtual machine (VM).
We’re pleased to say that Red Hat Virtualization will be the first hypervisor in the industry to support these multiple GPU workflows from NVIDIA. By collaborating closely with NVIDIA, Red Hat plans to make the multiple vGPU technology available through its open, enterprise-ready virtualization platform, making it easier to use NVIDIA GPUs to drive performance and innovations across a wide range of industries.
In addition to Red Hat’s contributions to upstream projects focused on Linux kernel and Kernel-based virtual machine (KVM), we have been working with NVIDIA and others in the upstream Linux community to enable support for mediated device framework (mdev). The framework manages device drivers calls enabling applications to take advantage of the underlying hardware. NVIDIA GPUs is one of the device types that mdev targets and support for this functionality is slated for inclusion in Red Hat Virtualization 4.2, a KVM-powered virtualization platform, in the coming weeks.
Now industries that depend on GPU accelerated applications, such as oil and gas, media and entertainment, automotive, healthcare, academic research and education can use this capability to deploy compute and graphic-intensive applications in a consolidated and virtualized environment, harnessing the benefits of enhanced computational and performance gains offered by NVIDIA GPUs.
Organizations can further lower IT costs and gain improved efficiencies offered by Red Hat Virtualization by providing their users with access to virtual computing resources that offer performance similar to a physical desktop or workstation.