Nvidia's CEO Jen-Hsun Huang has opened this year's GPU Technology conference with a keynote detailing their offerings, achievements and future plans. Most topics covered in it were related to GPU compute and professional design applications, but few topics are interesting for gaming community as well. Among those were several nice looking tech demos and an updated GPU roadmap.
Tech demos are used by GPU manufacturers to show what rendering techniques their latest offerings are capable of. For the most users, however, they are just nice looking visual showcases, often giving a few elements level of detail far exceeding the games of the time, which have to use graphical resources for other things as well. It seems that with Kepler GPUs, Nvidia has decided to revisit ideas shown in older demos and use the increased graphical power to give more detail. Last year we have received an updated vision of older Dawn demo and now we are looking at another bald head.
While it might seem that an another lone bald head shouldn't be too impressive these days after the in-game details of Crysis 3 and TressFX, there is still a lot of room for more realistic human rendering. Ira tech demo uses the performance of GTX Titan to provide very realistic skin rendering with its complex translucencies and sub-surface scattering. Most importantly, in collaboration with Dr. Paul Debevec's team from ITC USC, Nvidia used FaceWork technology to animate the Ira based on high quality pre-recorded human facial expressions. Nvidia claims that Ira has overcome the "uncanny valley", but you can decide that yourselves.
Another interesting demo was WaveWorks demo, showing an ocean simulation based on wind conditions. It wasn't as visually impressive as Ira, but its main aim was the simulation, not the graphical part.
Finally, Nvidia has shown an updated GPU roadmap showing Kepler's successors and planned features. Kepler based cards will remain the main offering throughout the 2013, with Maxwell planned for 2014 release. Among the expected performance improvement, the new architectural feature is memory sharing between CPU and GPU. GPU will be able to access main system memory directly, allowing it use a larger memory pool when the speed is less important than the amount. CPU will have the direct access to GPU's GDDR5 memory as well. For those interested, AMD also has plans to implement similar functionality in 2014. Continuing with their scientists naming scheme, Maxwell's successor will be Volta GPU. For it, Nvidia plans to stack VRAM directly on top of GPU, improving bandwidth to 1 TB/s range (Titan, 7970 have 288 GB/s for comparison). No planned date was named for Volta, but if trend continues we can expect it around 2016.
