The graphics processing unit (GPU) has evolved from silicon that only gamers cared about to something that's now widely used for accelerating power-intensive applications. GPUs are now important for machine learning, design and visualization, and data analytics. One of the challenges for many organizations is accessibility to GPU-enabled systems because the initial cost can be quite high and many organizations do not have the technical skill to deploy the hardware and software necessary to take full advantage of the technology.

Also, due to the pandemic, there has been a trend to remote work that has now evolved to many businesses having employees permanently work from home. This is fairly simple for knowledge workers, call center agents, salespeople, and others who work with software. But for data scientists, graphic designers, engineers, and others who need access to GPU-enabled systems, working remotely means no access to the systems they will need to do their jobs. 

Featured

• We're not ready for the impact of generative AI on elections
• This is the$300 Android phone to beat in 2023 - and it even has a stylus
• 5 things I learned while building my smart home
• The best laptops under$1,000: MacBook, Surface Pro, HP models compared

To help organizations obtain remote access to GPU-enabled systems, at a lower cost, with less complexity, NVIDIA has partnered with the major cloud providers to offer cloud-based virtual workstations. This has become particularly important since the start of the pandemic because IT professionals, data scientists, and others who need access to GPU systems are now working from home and can't easily connect to the systems located on company premises. NVIDIA has assisted customers in making this transition by offering free virtual GPU (vGPU) software trial licenses to get started.

Virtual workstations have been works in progress

Historically, virtual workstations have not performed as well as the physical ones, but that was because GPUs were not part of the mix. Customers deploying VDI in on-premises data centers can access GPU-accelerated performance with NVIDIA's vGPU technology. NVIDIA vGPU includes both an NVIDIA GPU and an NVIDIA vGPU software license. The NVIDIA A40 GPU has 48 GB of frame buffer and supports NVIDIA RTX technology -one of NVIDIA's most important advances in computer graphics -making it possible for creative and technical professionals to simulate the physical world at unprecedented speeds. 

With the A40 combined with NVIDIA RTX Virtual Workstation (vWS) software, remote workers can experience performance indistinguishable from a physical workstation one might find on an engineer's desk. For knowledge workers accessing office productivity applications, NVIDIA Virtual PC (vPC) software paired with NVIDIA A16 GPUs is designed to maximize user density per GPU as well as deliver the highest quality user experience. By pairing NVIDIA's vGPU software, such as NVIDIA vWS or vPC with NVIDIA GPUs, we create vGPUs that can be shared across multiple users. 

To get an idea of how effective vGPUs are, I recently conducted several scenarios to test the vGPU and compared them to CPU-only systems. The system I tested was the NVIDIA A16 Ampere card with NVIDIA Virtual PC (vPC) software; then the A40 paired with NVIDIA RTX vWS software. Finally, I compared identical workloads on CPU-only systems. 

Following are the results: 

Use case No. 1: OpenGL with NVIDIA vPC

Open Graphics Library, more commonly known as OpenGL, is a processor-intensive API to render 2D and 3D vector graphics. To test how this worked GPU accelerated, I went to an OpenGL website and noticed it loaded very fast and smoothly. The site had a number of animations built into it, and when I clicked to start one, they started almost immediately and continued to run smoothly. As I added more animations, there was more movement across the screen with no degradation in quality. When I ran the animations with CPUs only, the more animations I added the poorer the quality to the point where the graphics were barely moving. Instead of being smooth like the vGPU, the animation was painfully slow. 

Use case No. 2: YouTube in a virtual desktop accelerated by NVIDIA vPC

I then proceeded to run a YouTube video on a virtual desktop to see if the playing of the video would impact the processing capability, and it did not. Another notable point is that the sound and video remained in sync with the video with no blurring. When a CPU-only processor is taxed, the voice and video would quickly get out of sync and then start to buffer.  The performance difference wasn't as dramatic as the OpenGL use case, but it was noticeably worse on the non-GPU system. 

Use case No. 3: Large Microsoft Excel manipulation on NVIDIA vPC

The next test was to open a large Excel file, insert a 3D chart, and then rotate it to view the data in a graphical format. I could rotate the chart back and forth and visualize the data from a number of different views. With a CPU-only system, the 3D chart wasn't nearly as nimble to move around. After a few minutes, the time it took to have the graphic catch up made chart manipulation unusable. 

Use case No. 4: SolidWorks Visualize accelerated by NVIDIA RTX vWS

To give the new A40 a good test drive, I loaded a SolidWorks file. I loaded the image of a motorcycle. I then performed a number of tasks, such as rotating the image and zooming in and out. SolidWorks uses ray tracing with quick set rendering to redraw the image. I then proceeded to change paint colors, textures, and other attributes of the bike and saw no degradation in performance. Finally, I changed the background scenes from being predominantly black to mountains, an old warehouse, and other graphically intensive images. Again, performance remained constant. What was most impressive is the image was very lifelike and one could see a reflection of the bike on the wet floor and other things to make it photo-realistic. One of the benefits of using SolidWorks is the tool provides some data on time to re-render the image. For the vGPU, this time was barely a second or two.

I ran the same simulations with a CPU-only system and the time to render jumped to 30 seconds or more. It was very difficult to pan, zoom or rotate the image because it took so long to redraw the bike. The same was true for any color change, texture, or background. 

NVIDIA RTX Virtual Workstations in the cloud are a great way for businesses to get started with accelerated computing because the cost of entry is nominal compared to having to purchase individual physical workstations. Once utilization hits a certain volume, it may make sense to purchase NVIDIA virtual GPU software licenses and run them on-premises in GPU-accelerated servers. 

With the world moving to remote work, vGPUs may become the norm.