Advanced Computing in the Age of AI | Friday, March 29, 2024

NVIDIA CEO Jen-Hsun Huang Takes Design to the Grid at GTC 

<img style="float: left;" src="http://media2.hpcwire.com/dmr/gtckeynote1.jpg" alt="" width="95" height="75" />Whether companies use big data for quality testing, product tracking or price optimization, data volumes are growing to unprecedented levels for major organizations and small to medium-sized manufacturers (SMMs) alike. Hoping to address this issue is NVIDIA CEO Jen-Hsun Huang, who addressed solutions for organizations of all sizes in opening keynote at the GPU Technology Conference in San Jose, Calif., on Tuesday.

As manufacturing giants such as General Electric have already begun to show, big data has broken into the manufacturing sector. Whether companies use it for quality testing, product tracking or price optimization, data volumes are growing to unprecedented levels for major organizations and small to medium-sized manufacturers (SMMs) alike. This means that allowing employees across the enterprise access to the same information in real-time is becoming problematic.

This is just one of the topics that NVIDIA CEO Jen-Hsun Huang addressed in his opening keynote at the GPU Technology Conference in San Jose, Calif., on Tuesday.

Huang discussed five topics that address the future of GPUs: next-generation graphics, GPU computing, the road map for their upcoming Volta and Parker processors, remote graphics for enterprise computing, and what Huang described as the “world's first visual computing appliance.”

As Huang explained, the growing trend of “bring your own device” (BYOD) or the “consumerization of IT” has changed the structure of enterprise networks from a set of homogenous collection of company-issued workstations. As company computers become a thing of the past, employees are bringing their own laptops and tablets to create a heterogeneous environment.

Huang pointed out that copying files from the server to process and store it locally made sense when data volumes were small. Files could then be uploaded back to the server, giving the rest of the organization access to the new data. Despite appearances, this was the antithesis of staying in sync.

“It makes more sense to move the processing to the data, rather than bringing data to the processing,” Huang concluded. He envisions virtual GPUs as the solution to this problem, as they enable users to generate graphics using any PC or mobile device, thanks to the processing power of a server just behind the scene.

To demonstrate, Huang pointed to the example of Applied Materials, a semiconductor maker that houses some of the most complex manufacturing equipment around. Despite the products' small size, nanomanufacturing requires room-sized equipment, with thousands of engineers working on this simultaneously.

Here, synchronizing data between each workstation would have been an impossible task, so Applied Materials has hosted its desktops and applications within datacenters. The result is that employees can connect from any computing device to access server CPUs/GPUs, and applications are run on powerful servers connected to enterprise data via hyperLAN.

For large scale enterprise applications such as this one, Huang pointed to NVIDIA's GRID enterprise servers as a powerful tool for providing graphics processing to a number of users in a synchronous manner.

Next -- A Solution for SMBs >

A Solution for SMBs

But some people's work can't be solved this way, which is what brought Huang to the final topic of his keynote address. He pointed out that SMBs (or SMMs) don't have an IT department, despite having the same computational challenges and the same desires to work on one massive database from anywhere on any device. Furthermore, because small businesses often work with data belonging to a client or partner, security is paramount.

Rather than having a large rack of enterprise servers, Huang says that the solution is much simpler: what NVIDIA calls the first “Visual Computing Appliance,” or GRID VCA. He stressed that this is not a server, but NVIDIA's first system. Sitting only 4U in height, the appliance fits into a server rack. Inside are two 16-thread Xeon processors with 392 gigabytes of system memory per processor. They are backed by eight GRID GPUs, each with two Kepler GPUs.

The GRID VCA was demonstrated with a Macbook Pro. Taking advantage of three of the VCA's possible workstations were Autodesk's 3D Studio Max, Adobe Premiere and SolidWorks. While an unknowing viewer might assume these were all using the Mac for processing power, 3D Studio Max is not OS X compatible, while Premiere would not be capable of such real-time image processing given the laptop's form factor.

Helping Huang to preview SolidWorks was Gian Paolo Bassi, vice president of research and development at SolidWorks. Present in over 80 countries, about 180,000 different companies and 800 solution partners, there is no doubt as to the company's scope. However what may be more surprising is that SMMs instead of manufacturing giants make up the majority of their business.

Although anything from medical, automotive and aerospace equipment to consumer products has been produced using this 3D CAD software offered by Dessault Systèmes, on screen was the 3D model of the world's fastest motorcycle. Created by Lightning Motorcycles 2006, it made its first landspeed record in 2009, and in 2011 became the first electric bike to break 200mph, leading the company to contend that they have the most aerodynamic motorcycle ever built.

According to Bassi, GRID is so essential to customers like Lightning Motorcycles because it gives them access to the data and programs they need at any time, allowing them to review and tweak fundamental design elements whenever necessary. While mobile computing is already used by many of SolidWorks' customers, real-time graphics rendering made possible by GRID will take the experience to the next level, especially in the case of SMMs.

Huang added that the GRID VCA also accommodates the companies that have a workforce that constantly is changing its size. This means that IT departments no longer have to download necessary data and applications onto a new employee's machine since it's already available to them via GRID. For companies with design offices located throughout the world, this appliance could help solve the deployment problem.

Although the appliance is capable of running 16 virtual machines, any number of users can access them thanks to unlimited software licensing. This means that dozens of engineers from a single SMM can have their own powerful workstation while only requiring the company to invest in a single piece of equipment. The system demonstrated at GTC is priced at $39,900, while an additional model will feature half the specs at $24,900. The GRID VCA is currently in beta.  

EnterpriseAI