DaaS brings GPU-accelerated applications within reach

As desktop-as-a-service offerings step into the spotlight, organizations may realize the benefits of GPU-accelerated applications in cloud-hosted environments.

IT can deploy virtual desktops on premises, purchasing the physical hardware and GPUs to host in its own data center. Alternatively, an organization can deploy desktops with GPU-powered applications hosted in the cloud, in which a service provider hosts the GPU infrastructure, data and VDI in a public cloud.

The hosted desktop model will likely make GPUs more attainable for many organizations running virtual desktops and apps, said Mark Bowker, senior analyst at Enterprise Strategy Group in Milford, Mass.

"Companies can say, 'You know what, we're tired of having to buy the costly desktops and laptops with GPU chipsets inside; we're tired of having to refresh them on a frequent basis,'" Bowker said. "'We're going to take that completely off the table.'"

On-premises VDI vs. DaaS

GPUs allow for large chunks of data to process more quickly than with CPUs. Organizations that use graphic-intensive applications, such as those in the manufacturing or engineering industries, can use GPUs to process data from computer-aided design or imaging applications, for example.

Mark Bowker

It is possible for organizations to deploy GPUs on premises through providers such as Citrix Virtual Apps and Desktops and VMware Horizon, but not without a potentially complex VDI deployment, Bowker said. Cloud-hosted desktops offload many management tasks to the vendor and can offer a more cost-effective, predictable way to access expensive GPU hardware.

Deploying cloud desktops was necessary for Dudek, an environmental engineering firm in Encinitas, Calif., to process drone flight data.

"[On-prem VDI] is great if you can set it up right ... but you need someone to stay on top of it and make sure all of those components are running properly," said Dudek CIO Brian Nordmann. "We were looking for a turnkey solution ... so IT staff wouldn't be bogged down with babysitting systems."

The organization previously ran Citrix XenApp, but it experienced issues with an undersized network and poor performance over the WAN that caused data processing to take up to five days. As part of a cloud transition, Dudek purchased Workspot GPU cloud workstations, which offer preconfigured virtual desktops hosted in Microsoft Azure that can instantly access Nvidia GPUs.

"We needed GPU processing close to the data," Nordmann said. "That would be expensive to do ourselves with on-prem."

Brian Nordmann

The success of other desktop-as-a-service (DaaS) offerings such as Amazon WorkSpaces and the announcement of Microsoft Windows Virtual Desktop will likely give more validity to cloud-hosted desktops over on-premises VDI, Bowker said. Windows Virtual Desktop can support applications that require GPUs, but IT must first make an Azure VM that works with GPUs, add it to the host pool and configure it properly. Amazon WorkSpaces offers GraphicsPro bundles that come with a dedicated Nvidia Tesla GPU.

"The bar ends up being, 'Can you provide a bare-metal-like experience in a hosted desktop environment in a highly graphic-intensive use case?' [DaaS providers] are all positioning to say, 'Yes, we can already do that.'" Bowker said.

Benefits of hosted desktops for GPUs

The cost-effectiveness and predictability of cloud is a significant draw for organizations that don't have the option to invest in GPU hardware upfront. Workspot customers, for example, pay a flat annual fee to access the virtual desktops through its platform as a service. The subscription also includes the costs of hosting and managing the GPU infrastructure components in Azure.

[On-prem VDI] is great if you can set it up right ... but you need someone to stay on top of it and make sure all of those components are running properly.

"Once you make a capital investment, that's in the books for years," Nordmann said. "Going through a subscription, you don't have that sort of risk."

The DaaS cost model also makes it easier for organizations to scale when their needs for GPU-powered apps change. Dudek, for instance, needed to limit the risks of a potentially volatile drone imaging industry and be able to quickly add desktops with GPUs using the golden image when the organization agrees to another project, Nordmann said.

Plus, innovation often happens more quickly with GPU-accelerated hardware, and the devices can experience lifecycles of two to three years, as opposed to five to seven years, Bowker said.

"If a new GPU chipset comes out, I can start to initially take advantage of that as soon as the cloud provider supports it, as opposed to me having to make a new investment on-prem," he said.

App virtualization and GPUs

Sometimes, it's enough to just virtualize particular applications in the cloud, especially if IT needs an application to perform a certain function without being tied to a specific operating system, Bowker said. One use case is the ability to provide access to video training via a browser, for example.

Application virtualization with GPU support is a good fit in the oil and gas industries that use specialized, graphic-intensive legacy applications such as those for well planning and geological monitoring. When users need to access these applications via the web, it's often expensive and time-consuming to rewrite them for a web-based interface, said Aaron Cooke, co-founder of Helio Summit, a management consulting firm in Austin, Texas.

A tool such as Cameyo, a cloud-based application virtualization vendor, can help deliver shared GPU-accelerated virtual applications. Organizations can reduce costs by supporting eight to 10 users per GPU instance with Cameyo, which can host the applications on Google Cloud Platform or in any environment an organization chooses. 

"I can just manage the application itself and then give access on an as-needed basis," Bowker said.