NVIDIA Tegra system-on-a-chip was first introduced in 2008, and targeted for the smartbook and mobile device market. Microsoft’s Zune HD was the first high-profile product that placed the Tegra chip on the spotlight but now four years later, NVIDIA’s latest generation Tegra 3 (codename “Kal-El”) SoC has found home in multiple tablets as Google’s official reference platform and in upcoming high-end AUDI in-vehicle infotainment implementations . The third generation 40nm Tegra SoC boasts a wealth of improvements over the second generation Tegra design and upcoming fully mapped future releases are aiming for as much as 75X faster performance compared to Tegra 2.
Two main concerns when designing Tegra involves power and frequency. Mobile devices require optimally lower power consumption for prolonged battery life while increasing demands for graphically rich and HD capable processors means a new approach and compromise to design must be made. NVIDIA has therefore designed the latest generation Tegra SoC with Variable Symmetric Multiprocessing technology which minimizes active standby-state power consumption and has the capability to flex quad-core performance when required. NVIDIA has done this by adding a 5th companion-core to the four main performance optimized cores. These two different CPU core types utilize the same ARM Cortex A9 architecture but are built on differing process technologies. The 5th companion core is built on low power process technology, operating up to a modest 500MHz. While this sounds low, this is more than enough for background tasks like synching, updates, audio and even video. Most of the video playback and encoding functions are handled by onboard hardware-based encoders as well as decoders anyway and does not require additional processing from the high-performance cores.
When they are not needed, the high-performance cores are power-gated off and all the tasks are handled solely by the companion core. Whenever users would engage in video conferencing or gaming however, the companion core shuts off and the high-performance cores kick in. These functions are all handled by the firmware and can dynamically switch between four different states: Single companion core, Single performance core, Dual performance-core and Quad performance-core operations.
Looking even closer, each core has a new Media Processing Engine, 32KB instruction cache as well as a 32KB data cache and all cores share a 1MB L2. For graphics, Tegra has a custom ultra-low power (ULP) GeForce Kal-El GPU with 12 cores, 8 of which are pixel shaders (2X as much compared to Tegra 2) and four vertex shaders. On Android devices, Tegra 3 brings full 3D Stereo function as well as support for OpenGL ES Version 2.0, OpenVG 1.1, and EGL 1.4. The memory interface remains at a 32-bit width but low-power DDR2 support has been increased to 1066 as well as adding DDR3-L 1500 support. Video and image signal processing has also been improved with a decoder that can output 1080p H.264 high-profile content at up to 40Mbps and an encoder that can handle 1080p H.264 base profile. Like Tegra 2, Tegra 3 can handle dual simultaneous display outputs but at a higher resolution up to 2048x1536 on an LCD.