Data scientists are working from home now, which is unusual and possibly detrimental to the field, according to a recent market intelligence report. Part of the reason that performance may be down is the fundamental interdisciplinary nature of data science.
Data scientists work interdepartmentally using hypervisors, which run multiple virtual machines as a host machine. Each worker on the interdepartmental team uses guest machines, and the hypervisor provides guest operating systems on each one with a virtual operating. The hypervisor also manages the performance of each guest operating system.
COVID-19 is also unusual as a data phenomenon, which could complicate things for data scientists using real-time information for analyzation and prediction. There is no prior event that is similar to COVID-19 in digitally recorded history. There is no digital data stream that stretches back to the last similar event, which might be the flu pandemic of 1918.
Data science uses huge sets of information that can take many forms. Sometimes it is in the form of words and numbers, like those found in Excel spreadsheets. However, new forms of data science have sprung up in recent years that require the heavy-duty parallel processing of GPUs. This is especially true for the area of image analytics.
When data scientists are using image sets, they must determine several important factors about how their algorithms are defined to know whether they are suitable for GPU implementation. For example, if the method of the algorithm is majority data parallel (meaning it can perform the same instructions on multiple data elements in parallel) and the thread count of the images is equal or greater to the number of voxels or pixels in the images, then the algorithm would benefit from GPU implementation.
If you are on a data science team that uses GPUs to process information, you would typically have them installed on workstations, laptops, or a server at work. COVID-19 forced most of the workforce to work remotely. And since the interdepartmental teams are in a sense distributed within a company’s offices, continuing data science work during COVID-19 presents additional challenges.
But NVIDIA has vGPU software that allows data scientists to use their NVIDIA GPUs located in their work offices remotely from home. The software was originally designed for the normal fractional amount of total remote work that a company might require. In light of the pandemic, the vGPU software has become more critical for continuing important data science work from home.
NVIDIA vGPU software is installed on the physical GPU of a company’s server and creates virtual GPUs. These virtual GPUs are shared between multiple virtual machines as directed by a company’s IT department. IT also has a special role to play with vGPU software because it defines GPU acceleration for different types of work on any computing device regardless of its physical location.
And there are different subsets within vGPU that make distributing resources on virtual machines easier to do. For simple hosting of desktop environments on a central server, NVIDIA GRID software can help virtualize applications and operating systems as needed.
For workers who use computing-intensive graphics apps, NVIDIA Quadro Virtual Data Center Workstation software helps distributed or remote workers collaborate in real time from any connected device, no matter the location.
And for the heavy-duty data science and artificial intelligence (AI) workflows, a subset of vGPU is NVIDIA Virtual Compute Server software. This application allows IT to specify GPU resources through server virtualization, enabling data scientists to perform their work on virtual machines. This type of server-oriented GPU virtualization and distribution of virtual machines is crucial for data scientists who are working from home due to COVID-19.
And recent research from IDC indicates that employee productivity was the principal concern that companies and organizations had about making the shift to more permanent remote work due to the pandemic. IDC recently interviewed NVIDIA customers using GPU-accelerated virtual desktops, and according to a recent blog post, “organizations with 500-1,000 users experienced a 13 percent increase in productivity, resulting in approximately more than $1 million in annual savings.”
Customers using GPU-accelerated virtual desktops can take advantage of vGPU for data science and AI workflows.
New Version of vGPU
NVIDIA just announced that a new version of vGPU will be released in July 2020. The new version of vGPU focuses on operational productivity for IT administrators and additional support for more workloads. This makes sense as they are the key players using vGPU features to allocate computing power for specific workflows on virtual machines.
The most relevant part of the new version of vGPU for data scientists is NVIDIA’s new partnership with SUSE for its Linux Enterprise Server. The partnership between the two companies will provide vGPU support on the kernel-based machine platform through NVIDIA vComputeServer software, expanding hypervisor support for data scientists who often use Linux.
Many of the new features of the latest version of vGPU help IT perform virtualization tasks with much greater efficiency.
Sometimes when allocating resources, there are interoperability issues caused by the host and guest vGPU software running on different versions, making upgrades and system-wide deployments challenging. The new version of vGPU has cross branch support, allowing IT teams to operate vGPU software when host and guest virtual machines use different versions of the software.
For larger data centers that use VMware vSphere, the new version will help streamline operations by managing vGPU virtual machines with the most recent version of VMware vRealize Operations.
Bottom Line
COVID-19 has caused potentially permanent changes in the workforce, with many companies rearranging managerial and operational productivity to reflect the new normal of remote work. Data science is a challenging area to transform into an efficient remote operation, due to its interdisciplinary nature and the large amount of GPU computing resources it requires. IT teams in charge of GPU virtualization need better and more ways to assign and allocate virtualization resources for different applications and workflows, which is precisely what vGPU does. The new version has some key features to help IT departments and data scientists improve operational efficiency as remote workers retain continuity as best they can. The interruption of COVID-19 continues, and with things on shaky ground, important data science work has to continue remotely.
The exact date for the new vGPU release is unknown at this time. NVIDIA has just said the release will be out sometime later this month.