Nvidia wants to bump up graphics and application performance on tablets and smartphones with the two new Tegra mobile processors.
The new Tegra processors will be called Logan and Parker, and will succeed the Tegra 4 processor expected to reach smartphones and tablets later this year. The new chips were announced as part of a road-map update provided by Nvidia CEO Jen-Hsun Huang during a keynote at Nvidia's GPU Technology Conference being held in Santa Clara, California.
Huang spent some time talking about major enhancements in the new chips, but did not provide detailed specifications such as processor speeds. Nvidia has a trend of naming Tegra chips after superheroes, and the Logan code name likely is based on a character in X-Men, while Parker could be a reference to Spiderman.
The Logan chip will be the size of a dime and the first to follow Tegra 4, Huang said. The first chip will likely become available later this year, though Huang indicated that mass manufacturing of the chips would begin next year.
The biggest enhancement in Logan is the inclusion of graphics cores based on the Kepler architecture, which will provide a big graphics performance boost to smartphones and tablets. The world's fastest supercomputer, called Titan and located at the U.S. Department of Energy's Oak Ridge National Laboratory, uses Nvidia's graphics processors based on the Kepler architecture. The supercomputer delivers 20 petaflops of peak performance.
Logan will also be the first Tegra chip to support CUDA for mobile processors, which will allow programmers to write applications that jointly harness the computing power of CPUs and GPUs. Logan will support CUDA 5, which is a set of programming tools offered by Nvidia for its graphics processors to develop and manage parallel task execution.
"Logan has something that we've been dying to bring to the world for so long," Huang said.
The follow-up to Logan will be Parker, which will be the company's first 64-bit Tegra processor. Parker will be based on ARM's 64-bit ARMv8 process architecture and Nvidia's chip design called Project Denver, which was announced two years ago.
The Parker chip will have Nvidia's upcoming graphics processor technology called Maxwell, which unifies the CPU and GPU memory. With the ability for GPU memory to read CPU memory and vice versa, developers may find it easier to write programs, Huang said.
Currently GPU and CPU memory are divided and based on different technologies, but can be theoretically linked via virtualization technologies. Linking them makes it easier for the processors to share multiple threads and ensures that workloads and their branches are handled and executed correctly.
The Parker chip will also have 3D transistors, in which transistors are stacked on top of each other. That is different from current chips in which transistors are arranged next to each other, also called the planar structure. The 3D structure, called FinFET by the semiconductor companies, usually yields performance improvements and power savings, which could help speed up smartphones and tablets while preserving battery life.
The 3D structure was first incorporated into Intel's chips based on the 22-nanometer process. Foundry companies that make ARM chips like TSMC (Taiwan Semiconductor Manufacturing Co.) and GlobalFoundries are in the process of incorporating technologies to manufacture chips with 3D transistors.
Huang did not provide a release date for Parker chips. However, ARM has said chips with its 64-bit processor architecture will reach devices around 2014.
The first Tegra 4 processors, based on ARM's Cortex-A15 design, will be used in a ZTE smartphone due to be released in China by midyear, and also will be used in Nvidia's portable handset called "Project Shield," which will become available in the second quarter this year. Nvidia has also announced a Tegra 4i chip, which has a modified Cortex-A9 processor core and an integrated software-defined LTE radio.