By Charles King, Pund-IT, Inc. March 25, 2015
As noted in last week’s Pund-IT Review, I attended NVIDIA’s recent GPU Technology Conference (GTC) 2015 in San Jose mainly for the OpenPOWER Foundation sessions. But a visit to the conference’s exhibit hall and discussions with other attendees got me thinking about just how far graphics processing technologies have progressed over the past decade. In fact, the news out of GTC 2015 suggests that businesses that ignore developments in this space do so at their peril.
Why so? Because NVIDIA and its partners, including those in OpenPOWER, along with competitors across the greater IT industry are migrating innovations once relegated to graphics professionals and gaming enthusiasts into increasingly common business processes and IT products. In other words, if sophisticated, graphics-rich solutions aren’t already in your and your competitors’ offices and shops, they likely will be soon.
GPUs evolve into the mainstream
It’s arguable that the spark for this migratory evolution began in 2002 when IBM began developing a supercomputer for the U.S. Department of Energy’s (DOE) National Nuclear Security Administration. Called “Roadrunner,” the system was the first “hybrid” supercomputer that leveraged both processors and co-processors or accelerators. In Roadrunner’s case, the system consisted of 6,480 AMD Opteron dual-core processors and 12,960 IBM PowerXCell 8 co-processors that combined the company’s Power architecture and the Cell GPU technologies IBM developed with Sony and Toshiba for use in Sony’s PS3 gaming consoles.
Roadrunner was designed so that serial and parallel processing tasks could be seamlessly shared or apportioned to fully maximize performance, and the results speak for themselves. In 2008, Roadrunner became the first supercomputer to break the petaflop performance barrier, achieving 1.026 petaflops shortly after its installation at the Los Alamos Lab and 1.456 petaflops later that year. During its lifetime, the system was used to model the aging of the U.S. nuclear arsenal, along with other tasks in the science, financial, automotive and aerospace industries.
Roadrunner led the Top500.org list of top performing supercomputers for three consecutive lists (June and November 2008 and June 2009), and remained in the top 10 through November 2011. Roadrunner was still one of the world’s top 30 supercomputers when it was decommissioned in 2013, replaced by more energy efficient technologies. But the system’s hybrid design opened new vistas for supercomputing and HPC innovation. In the most recent Top500 list, four of the top 10 supercomputers are hybrid designs, as are 75 of the listed installations. Two thirds of those hybrid systems (including #2 and #6 in the top 10) leverage NVIDIA GPUs as co-processors/accelerators.
What was new at GTC 2015?
Given that lead-in, it would be reasonable to assume that the rest of this commentary would be supercomputing-focused, but that would be only partly correct. While NVIDIA (as well as Intel with its Xeon Phi chips and AMD’s ATI Radeon silicon) has helped to expand the continuing commoditization of supercomputing and high performance computing (HPC), it has not been bound by that category alone. Among the many announcements the company made at GTC 2015 were:
- A close look at TITAN X, NVIDIA’s next generation GPU that delivers twice the performance and power efficiency of its predecessor by combining 3,072 processing cores with 12GB of onboard memory for 7 teraflops of peak single-precision performance. Along with taking gaming to the next level, NVIDIA said that TITAN X is an ideal platform for supporting deep learning processes related to artificial intelligence
- The coming availability of the DRIVE PX development platform that will help tier 1 auto manufacturers, suppliers and research institutions harness NVIDIA’s DRIVE PX self-driving car platform
- Presentations by Google, SpaceX, Pixar, pharmaceutical companies and the U.S Department of Defense which are using the modeling capabilities of NVIDIA GPU-based solutions to create new and next generation products.
That doesn’t include NVIDIA partner announcements or those made related to OpenPOWER. Of the former group, Dell’s launch of the industry’s first ever ISV-certified virtual workstation appliance solution, the Dell Precision Appliance for Wyse, may be the most immediately impactful for businesses. In essence, Dell is working closely with ISVs, like Siemens and PTC, and leveraging the NVIDIA Grid platform to manage and deliver massive data sets and support certified workstation-quality graphics performance across a range of endpoint devices, including Dell Wyse thin and zero clients. The resulting solution has numerous industrial use cases related to product design and development.
At OpenPOWER, NVIDIA’s GPU solutions played key roles in two high profile announcements. The first is collaborating with IBM, Wistron and Mellanox in planned supercomputing solutions that will be 5- to 10-times faster than today’s best performing systems. The second is a collaboration between NVIDIA and Tyan resulting in the immediate availability of the Cirrascale RM4950, a platform supporting the development of GPU-accelerated big data analytics, deep learning and scientific computing applications.
As an industry, IT has long been subject to and benefited from the cross fertilization of sometimes disparate, even incongruous technologies. That’s certainly the case in graphics processing which has evolved from being confined to costly solutions from specialist vendors like Silicon Graphics that drove state of the art computer generated imaging (CGI) special effects to becoming a part of everyday personal computing.
As is apparent from the rise of graphics co-processors and accelerators in supercomputing, the high end of the GPU market is alive and doing very well, indeed. But GTC 2015 also made it clear that NVIDIA and its partners, developers and customers have a far more expansive view of the roles and opportunities available for graphics technologies. The eye-popping, graphically-intensive solutions resulting from their efforts are likely to arrive at a nearby workplace far sooner than you might expect.
© 2015 Pund-IT, Inc. All rights reserved.