GPU Accelerated VDI
A little while ago, VDI expert and community fellow Ruben Spruijt (@rspruijt) asked Shawn Bass (@shawnbass) and me if we want to join him to write a new “smackdown” whitepaper on hardware accelerated graphics solutions for VDI. In the past, Ruben has published a number of such comparison whitepapers covering various topics, and they enjoy a great reputation in the market. The goal of this new whitepaper is to inform community and industry about use cases, vendors, solutions, technologies and differences around hardware accelerated desktop virtualization. But why is this important? This article sheds some light on why I truly believe that GPU accelerated virtual desktops and remote Windows applications will be a general trend in the near future.
Early version of Windows used function calls to the Graphics Device Interface (GDI) for drawing graphical objects and transmitting them to monitors and printers. In order to accelerate graphical output it was possible to use dedicated hardware for this purpose. Graphics cards with GDI support became very popular, providing the capabilities to perform the rendering of some graphical elements by the Graphics Processing Unit (GPU) and others by the CPU. Over time, all possible GDI functions were performed by the GPU while the price for such a feature built into graphics cards went down to literally zero. Today, almost all graphics cards support the full GDI function set.
When looking at video playback on mobile devices, it’s pretty much the same story. At the beginning, video was rendered by the CPU, resulting in less than ideal performance. Later, H.264 decoder chips were added to high-end mobile phone and tablet hardware. Today, all mobile devices come with hardware decoders for video streams. So hardware acceleration history repeats when there is enough demand.
The same is true for remoting into Windows applications. (And full desktops to a certain extent.) Using physical GPUs to improve graphics performance in VDI and remote application scenarios is still a niche. Most people think that use cases are limited to high-end 3D designer and engineer workplaces, but recently growth is driven by mobile devices, games and the use of better graphics in a variety of applications. New use cases for power users and knowledge workers get established in fields like CAD/CAM/CAE, automotive, aerospace, shipbuilding, architecture, financial services, utilities, geographic information systems, healthcare, and media/video production. It is important to note that due to the introduction of (private and public) cloud concepts and a growing number of mobile devices, hardware accelerated remote applications are increasingly used for editing purposes and not only for creating new material. I personally refer to this as “lite touch” use cases, sitting right between the pure creation and consumption uses cases on either side of the spectrum. Lite touch use cases and media rich applications suddenly make GPU accelerated remoting concepts even attractive for market sectors that traditionally didn’t introduce such a technology for their average users, such as education, manufacturing, construction, retail sales, agriculture, mining or forestry.
In addition, it turns out that the newest implementations of Microsoft RemoteFX, Citrix HDX 3D Pro or VMware vSGA/vDGA allow standard applications such as Microsoft Office and Adobe Photoshop to benefit significantly from using physical GPUs on the host side. All this together shows the huge potential GPU accelerated graphics remoting has. And pretty much like in the old GDI times, prices for graphics cards will drop significantly over the next years, making hardware accelerated application remoting a commodity. In the near future, not only CPU cycles, memory, IOPS and disk space will be offered by public cloud providers and on-premises datacenters, but also GPU resources.