Nvidia’s GTC: Making the imaginary real

Nvidia’s GPU conference has become one of the more interesting industry events as the graphics market scales up to supercomputing and down to in-car navigation and entertainment systems.

What started out as a company that was mostly focused on gaming and CAD has evolved to create a broad event with speakers that range from DNA scientists to CEOs of emerging companies in developing nations.

As with all events, the CEO keynote sets the tone of this event and what follows are my impressions from the keynote.

GPUs Make it Happen, What You Make is Possible

The opening appeared to convey the idea that a graphics engine is the enabler which results in a massive amount of change in a huge number of industries. The opening images had gaming at the center, but ultimately branched out to movie special effects, sound and automotive systems, and scientific research at a molecular level.

The start was pretty rough because there was clearly a lack of rehearsal which detracted significantly from the presentation. Pacing was off, and support was having difficulty showing the right image because the presentation changed in real-time. It was interesting to contrast this with the HP CEO presentation a few weeks ago, which was the polar opposite – poor content but well-rehearsed. The Nvidia keynote actually had far better content but the lack of rehearsal made it somewhat more painful to watch.


As with all shows like this, Nvidia opened with a product that goes back to its graphics roots: GeForce GTX Titan. The graphics card is stunning to look at but the focus quickly shifted to what the card can do. Ocean simulation was used as the example. This was pretty impressive, as the Nvidia team was able to create a real time simulation of a ship in an increasingly stormy night. White water, spray, the physics of the ship being pounded and the power of the storm were so realistic I pretty much gave up ever going out into the ocean again on a ship.

They then shifted to face rendering. Nvidia CEO Jen-Hsun Huang spoke about the uncanny valley, probably should be creepy valley, because at some point the rendered face becomes so real that it becomes creepy – where only the subtlest of differences are noticeable. However, we still notice them and the result, instead of being better, is actually worse because it is repelling on many levels.

Crossing over that creepy level has been a huge problem, particularly for movie makers who would like to be able to create real images in real time and, up till now, getting close enough has meant weeks of rendering farm work and even then the result often was

only good in fast action shots because people were distracted by the action and wouldn’t be creeped out.

They then showcased how far they have come by first showing an attractive but unrealistic scantily clad female fairy and the Titan rendered floating bald headed male. The realism improvement was dramatic though I kind of wonder about the decision to move from a hot girl to a bald floating head. Titan appears to be a very powerful graphics card but I found myself wishing they’d put the hot fairy back on the screen.

GPU Computing

This capability has expanded to a point that there are 50 supercomputers currently running Nvidia’s platform for GPU computing. GPU computing, which started as a small fraction of supercomputer work in 2008, now dominates the growth side of this industry .

The company showcased the Piz Daint supercomputer by Cray as one of the most advanced and designed to model the weather in Europe. Jen-Hsun moved on to talk about how this technology is used to model High Energy Physics, 3D Genomics, simulate materials, do medical research for things like Alzheimer’s disease, and conduct

chemical analysis. Other uses ranged from matchmaking to more efficient diamond cutting.

One of the areas of growing analysis is data analytics. Companies like Saleforce.com are analyzing up to 500M tweets in less than five minutes using GPU computing to develop results that executives can use to make better decisions about their marketing and sales strategies.

Shazam’s CTO was then brought on stage to talk about how they use GPU computing to identify songs through music search. They are currently running 300M music searches per month. They then showcased Cortexica, a company that can take a picture of something like a dress and then find a similar product on a service like eBay. What

is interesting is that it finds clothing that has similar colors, similar types of images (like flowers), and similar image styles (like Asian).


On the graphics card side, based on a technology named Cuda, they had and will have: Tesla, Fermi, Kepler, Maxwell, and finally Volta. Volta is their next generation of graphics processor which will process one terabyte of information per second. For perspective – that is the equivalent of 50 Blu-Ray DVDs per second.

Tegra is their mobile platform, and Nvidia is currently shipping Tegra 4 SoCs to OEMs for products that will show up later in the year. Next generation code named Logan, Tegra 5, introduces Cuda to mobile processors bridging the desktop and mobile families.

Following Logan is Parker using FinFET transistors and bringing Maxwell level technology to mobile devices. Logan is due to market in 2013 and Parker in 2015. Oh, and yes, Nvidia showcased a proof of concept dubbed “Kayla” running a Tegra 3 with a CUDA graphics system which could do smoke, water rendering, and ray tracing simultaneously.

Grid Enterprise

One of Nvidia’s major initiatives is their Grid platform. This is a centralized graphics resource which provides high powered graphics resources to remote workers, dynamically assigning the needed performance to the remote device. Virtually all of the major server makers have products they have released using this capability. This allows for better collaboration, better security, and apparently a far more cost effective way to address the market need for graphics performance.

At the event, Nvidia announced the A Grid VCA, a new integrated rack mounted graphics solution. It has 8 graphics cards and will support 16 virtual sessions. The demonstration connected a MacBook to this service showcasing a level of performance that the laptop alone just isn’t capable of.

Wrapping Up:

This event increasingly showcases the strength of Nvidia, and while the lack of rehearsal did detract from the impression created, there is no doubt that Nvidia is a power to be reckoned with. Much of what we will see in the future will likely come from their technology as we increasingly struggle to differentiate between what is real and what has been created from someone’s imagination. Essentially, that’s the power of Nvidia, making the imaginary appear real.