updated 08:06 am EDT, Tue August 12, 2014
Tegra K1 with dual-core 64-bit Denver processor, 192-core Kepler GPU due later this year
Nvidia has revealed some of the architectural details of the 64-bit version of the Tegra K1 System on Chip (SoC). Revealed eight months after the original 32-bit version, the 64-bit version uses Denver, an architecture project originally meant for use in servers, with the company claiming it will be the first 64-bit ARM processor for use with Android.
The SoC consists of the custom-designed, 64-bit, ARMv8 architecture-compatible, dual-core Denver CPU clocked at up to 2.5GHz, combined with the existing Kepler-based 192-core GPU already used in the 32-bit Tegra K1. Each Denver core uses a seven-way superscalar microarchitecture, capable of up to seven concurrent micro-ops per clock, along with a 128KB four-way L1 instruction cache, a 64KB four-way L1 data cache, and a 2MB 16-way L2 cache shared between both cores.
A process called Dynamic Code Optimization will apparently optimize frequently-used software routines at runtime, turning them into "highly-tuned microcode-equivalent routines" stored in a dedicated 128MB main-memory-based optimization cache. By storing and refetching the optimized code when required, this reduces the need to reoptimize routines, cutting down on processor load and effectively doubling the base-level hardware performance.
Device producers will benefit from the Denver-based Tegra K1's full pin compatibility with the 32-bit Tegra K1, simplifying the implementation and making it easier to bring products using the SoC to market. The first devices using the 64-bit Tegra K1 will apparently ship later this year, according to Nvidia, with the company also promising support for Android L in the future.