Nvidia officials, eight months after launching its 32-bit Tegra K1 mobile chip, is giving details about the upcoming version that will include a custom-designed ARM-based CPU and bring PC capabilities to such mobile devices as tablets.
At the Hot Chips 2014 conference this week in San Jose, Calif., Nvidia officials said the 64-bit Tegra K1 is the first 64-bit ARM processor for Android devices, pairing the dual-core "Project Denver" CPU with Nvidia's 192-core Kepler GPU. The Denver CPU was custom designed by Nvidia and is compatible with ARM's 64-bit ARMv8-A architecture.
It also is pin-compatible with the 32-bit Tegra K1 chip, enabling programmers to develop software that will run on either chip without any tweaking needed. For Nvidia, it's all about performance with the 64-bit version, according to Nick Stam, senior technical marketing director for the company.
"With its exceptional performance and superior energy efficiency, the 64-bit Tegra K1 is the world's first 64-bit ARM processor for Android, and completely outpaces other ARM-based mobile processors," Nick Stam, senior technical marketing director for Nvidia, said in a post on the company blog. "Denver is designed for the highest single-core CPU throughput, and also delivers industry-leading dual-core performance."
The 32-bit version has found its way into a number of devices, including Nvidia's Shield tablet for gamers and a new 13.3-inch Chromebook announced by Acer Aug. 12. The 64-bit Tegra K1 is due to launch later this year.
ARM is licensing its 64-bit ARMv8-A architecture to systems makers and is working on the next-generation version. Much of the focus of the 64-bit architecture has been on the data center, where ARM officials believe their highly energy-efficient designs—which can be found in most tablets and smartphones—can be leveraged in servers in such environments as clouds and hyperscale data centers. A range of chip makers—from Advanced Micro Devices and Applied Micro to Cavium and Marvell Technologies—are developing ARM-based server systems-on-a-chip (SoC).
During the Hot Chips show, Nvidia officials talking about the 64-bit Tegra K1 focused on what the chip can do in mobile devices. According to the company, the 64-bit Denver cores will offer seven-way superscalar architectures, which can execute up to seven concurrent micro operations per clock cycle, and can run up to 2.5GHz. By comparison, the 32-bit version—with a quad-core ARM CPU—offers a three-way superscalar architecture and runs at 2.3GHz.
In addition, Nvidia also is using a process called Dynamic Code Optimization, which takes ARM software code that is frequently used, makes it optimized micro-code and houses it in a dedicated 128MB main-memory-based optimization cache. It improves the chip's performance by ensuring that the frequently-used code is ready when needed, rather than the chip having to reoptimize it whenever it's needed, according to Stam.
"Denver extracts the ILP once via software techniques, and then executes those routines repeatedly, thus amortizing the cost of ILP [instruction-level parallelism] extraction over the many execution instances," he wrote.
It's reportedly similar to what long-gone Transmeta tried to do with the low-power chips it was developing more than a decade ago to challenge Intel.
"Denver also features new low-latency power-state transitions, in addition to extensive power-gating and dynamic voltage and clock scaling based on workloads," Stam wrote. "Combining Dynamic Code Optimization, seven-way superscalar design and efficient power usage, Denver's performance will rival some mainstream PC-class CPUs at significantly reduced power consumption. This means that future mobile devices using our 64-bit Tegra K1 chip can offer PC-class performance for standard apps, extended battery life and the best Web browsing experience—all while opening new possibilities for gaming, content creation and enterprise apps."