Wednesday, July 24, 2013

HARDWARE - nVidia's Next Generation Mobile GPU



"Kepler to go:  nVidia brings high-end graphics core technology to Tegra 5" by Agam Shah, PC World 7/24/2013

Augmented reality, image recognition and other multimedia features could be standard in future smartphones and tablets, and nVidia’s upcoming Tegra 5 mobile chip will have features to handle such demanding graphics capabilities.

nVidia on Wednesday said that it has made its biggest advance in mobile graphics technology with the integration of its latest graphics core code-named Kepler into Tegra 5, which is code-named Project Logan.  The chip is due next year, and will be able to handle the most demanding graphics applications through ray-tracing, tessellation, advanced lighting and post processing, said Daniel Vivoli, senior vice president at nVidia.

The graphics capabilities in Logan will be demonstrated at the SIGGRAPH show in Anaheim, California.  The demo will highlight the ability of a mobile processor to show a lifelike human face while consuming just two to three watts of power.  The 3D simulation of the human face will show “full features,” Vivoli said, including light refraction, microscopic wrinkles on the skin, and other small details such as skin oils.

The human face — called Ira by nVidia — was demonstrated on stage at nVidia’s GPU Technology Conference and was rendered with server-class graphics processors based on Kepler.  Features from those high-end GPUs are being scaled down to fit into the power constraints of mobile devices, Vivoli said.  The Ira demonstration was ported to Logan after paring down some hardware capabilities and also with tweaks in clock gating and cache.

Tegra 5 is scheduled to ship next year.  nVidia has just started shipping its Tegra 4 chip, which will be in devices such as Hewlett-Packard’s SlateBook X2 tablet.

nVidia declined to provide numbers on the graphics performance gains versus Tegra 4.  But the graphics core will be faster and more power efficient, and nVidia said it will use less than a third of the power of graphics cores in tablets like the iPad when rendering the same graphics.  Logan will provide better graphics performance at the same power consumption levels.

nVidia is known for its graphics, and its chips are considered among the best at handling multimedia in mobile devices.  The company’s high-end Tesla graphics chips based on Kepler are being used in some of the world’s fastest supercomputers, and now similar features will be available in mobile devices going ahead.

It’s also the first time that nVidia is bringing its latest graphics core to the mobile processor, effectively uniting all graphics products on the same microarchitecture.

“We’ve always have a separate architecture,” Vivoli said.  “We’ve been working for years where we can converge the graphics roadmaps.”

nVidia earlier offered a graphics development platform called Kayla in which a Tegra processor was attached to a Kepler GPU via a PCI-Express interconnect.  The platform was intended to get programmers to start writing mobile applications for the Kepler GPU.  But with Logan, the Kepler graphics processor is integrated inside the Tegra chip.

Programmers will have to write algorithms and programs to enable augmented reality, face recognition and other high-end multimedia, Vivoli said.  Processing such tasks will be quicker when off-loaded to the Kepler graphics core, Vivoli said.

It will also be the “first time” GPGPU (general-purpose graphics processing unit computing) comes to mobile devices, Vivoli said, referring to a concept in which processing is being increasingly moved to from CPUs to graphics cores in systems.

But the CPUs and graphics processors still need to work in a coherent manner, and the Tegra 5 chip will support a range of parallel programming tools such as CUDA 5.5, OpenCL 2.0 and Microsoft’s DirectX.  Such tools harness the joint processing power of CPUs and GPUs to bring performance gains in supercomputers, and with mobile devices, the performance boosts have to fit within a specific power limit.

There are multiple parallel programming development tools for mobile devices and supercomputing.  Intel offers development tools to work with its Xeon Phi accelerator chip, while Advanced Micro Devices is pushing with specifications from the HSA (Heterogeneous System Architecture) Foundation, a group that hopes to provide tools so applications can be easily ported across different chip architectures and devices.  nVidia is not a member of HSA, which is backed by ARM, Qualcomm, Texas Instruments and others.

Beyond Logan, nVidia is making more hardware improvements that should make graphics rendering faster.  The Tegra 6 processor code-named Parker will unite CPU and GPU and make it a shared resource.  Parker will also have a 3D structure in which transistors will be stacked on top of each other, which should make the GPU faster and more power efficient.  Parker is due for release in 2015.

No comments: