NVIDIA’s Greg Estes on Visual Computing Workflows


NVIDIA V.P. of Enterprise Marketing Greg Estes joins Steve Waskul for a visual computing workflow discussion covering the newest products being offered by NVIDIA.

The interview begins with Greg providing an overview of the next generation of NVIDIA Quadro GPUs that were just announced. The products include the K5200, K4200, K2200, K620 and K420 (The K6000 still remains the NVIDIA’s most powerful pro graphics solution and was not refreshed). Greg explains that on average the new products are about 40% faster than any previous generation under real-world application use. Some processes may run faster than this while other may be slower of course. In all cases the new solutions benefit from a doubling of GPU memory so the K5200 for example now features 8GB of GPU memory. This will help uses with larger data sets work more interactively at all product levels and that’s important. Greg uses the example of a Hollywood artist. If you can fit your entire model within the GPU memory then you can work with it interactively. If not, the data has to fetch back in and out which obviously slows the process down providing a user experience that is not truly interactive.

An interesting point Greg brings up is that NVIDIA did not think about just providing more performance at the same cost for this product refresh. While that’s always beneficial, he explains that they have been putting a lot of energy into taking with their high-end customers about what they need in their workflows. Companies like Pixar, Audi, GE Medical and Shell all need to deal with large data sets. If you are in Media and Entertainment you’re not only going from HD to 4K but now to 6K and beyond. You might also be working in Stereo or at a high frame rates as well. In the oil and gas arena you might have terabytes of information you need to look at for an oil field. In the areas of product design everything is getting more complex and even for firms where it isn’t necessarily more complex being able to speed up the design process can bring products to market faster. So, Greg explains that a “data explosion” is happening across all of NVIDIA’s enterprise customers and that’s why they’ve doubled the data capacity of all their Quadro solutions. With more graphics memory and more processing power you can work with twice as much data as you could before.

Another area that is common among many customers is that they are not sure how to integrate the cloud into their workflows according to Greg. Are all the applications supposed to be hosted on the cloud now? Is it just for “overflow”? How should it work? To address this, NVIDIA has optimized the new Quadro products for seamless interconnectivity to cloud-based resources. So, for example a Maya user could, with the new Quadro, just pull down a menu and connect directly to a cluster of NVIDIA Visual Computing Appliances without leaving the application. This is a powerful way to bring the performance of numerous NVIDIA high-end GPUs to one user. Because the cluster sits on the network, it is available to anyone on the network and it can be easily integrated into design workflows. Greg explains that the cloud just becomes part of the workflow.

The third thing that customers are looking for is Mobile. Younger folks are coming into the workforce and Greg says they want to bring their own device (BYOD) to work. They want to work on tablets. They want to do their own thing. It makes complete sense as mobile devices like tablets and laptops are getting more powerful people naturally want to incorporate them into the new workflow. NVIDIA is responding to this by incorporating their grid technology into the new Quadro. So, Quadro is at the center of the workflow and the NVIDIA GRID technology is bringing the rich user experience to the tablet or mobile device. The NVIDIA GRID technology does this through GPU virtualization and remote and session-management libraries so remote users can experience graphics-intensive desktops and applications remotely. Basically, Quadro is at the center allowing remote users to connect and get into the “Quadroness” as Greg calls it.

Next, the interview take a turn and Greg and Steve talk about the new feature film based on the Gillian Flynn’s best-selling novel Gone Girl. By most accounts this is truly a ground-breaking film. It is the first feature shot entirely in 6K and the first studio feature to be edited entirely in Adobe® Premiere® Pro CC. Greg explains that Director David Fincher built an innovative production pipeline for this project and that the Quadro K5200s were used for all stages of production. The 6K Imagery captured by DP Jeff Cronenweth on a RED Dragon camera was debayered directly in Adobe Premiere Pro. The Quadro 5200 allowed real-time downscaling and playback at 4K resolution eliminating the need for RED Rocket hardware.

Greg notes that its experiences like this that have colored NVIDIA’s thinking. They didn’t just come out with a whole new line of products and say they are faster and better at the same price. They are really thinking differently. One example he gives is the new Visual Computing Appliance. With 8 NVIDIA high-end CPUs and more than 23,000 CUDA cores, it is purpose built just to do GPU rendering. It works with both the NVIDIA® Iray® renderer and Chaos Group’s V-Ray RT which is the rendering engine of choice for design, broadcast and visual effects. V-Ray RT takes full advantage of the parallel processing capabilities of the Visaul Computing Appliance and is included with V-Ray for 3ds Max, V-Ray for Maya, V-Ray for Rhino and V-Ray for SketchUp. So, Greg says, this allows you to get fast photorealistic rendering within a wide range of tools.

Although we haven’t tested this solution at Waskul.TV, the concept is brilliant and gaining momentum as you can learn from other interviews on this site. As Greg notes for most design teams, how often does one really need maximum GPU power? Is it 20 minutes in a given day? 4 hours? Most would agree that it is typically not the entire day. That opens up the possibility for sharing GPU resources on the network and in this case opening up GPU speed potential that could not live within the workstation itself. Whether all the performance is dedicated to a single high-priority user or it is spread out evenly across multiple users in a workgroup, the point is that it is a network resource so you can get better utilization from it. Management is done through a simple web-based scheduling tool. The NVIDIA VCA also has 20 physical CPU Cores, 256GB of system RAM, 2TB of SSD storage and GbE, 10GbE and InfiniBand network connections.

Moving on Steve asks Greg about how NVIDIA works with workstation manufacturers and software companies to deliver optimized solutions for professional users. Greg explains that the collaboration between companies delivers significant value to professional customers because they can get a system that will be “locked down” for three to five years of use in a particular chassis with particular software. Actually, it turns out that a lot of the work his team does is working with workstation manufacturers and software developers to make sure they are delivering professional solutions customers can count on.

The final segment of the discussion focuses on how game developers work to optimize their games for all the different platforms available. Greg shares information here about NVIDIA’s programs aimed at helping developers in debugging / designing. Whether it’s a tablet or NVIDIA’s highest GPU, they have tools that help developers understand where resources are being used. Greg says that providing that level of debugging tools and the ecosystem, etc. is an important part of what NVIDIA does. He also mentions that in addition to CUDA, NVIDIA absolutely works with OpenCL and their solutions do a great job of accelerating OpenCL applications.