Adapteva has released the fourth generation of its Epiphany multicore architecture IP – a 28nm design with 64 independent high-performance RISC cores on a single chip. The latest generation of the Epiphany architecture offers the highest energy efficiency of any floating point processor solution to date, with a ground-breaking 70 GFLOPS/Watt.

“With the launch of the Epiphany architecture in 28nm, we’re able to bring even more computing power to a tiny footprint, without increased energy drain,” said Andreas Olofsson, CEO of Adapteva. “The Epiphany-IV platform demonstrates an immediate path to exceeding DARPA’s 2018 goal of 50 GFLOPS/Watt in high performance computing applications.”

CoreMark benchmarking data shows that the previously released Epiphany-III chip reaches equivalent performance levels of server type microprocessors like the Intel Xeon chips, while consuming less than 2W of peak power. The Epiphany-IV chip will be even more impressive with a 4X performance improvement on CoreMark benchmarks while keeping the power consumption the same as the previous generation. Adapteva’s low-power, high-performance approach will transform the mobile markets by enabling server-level computing local to portable devices such as smartphones and tablets.

Making Multicore User-Friendly

Among only a handful of companies offering 28nm chip designs, Adapteva is bringing parallel computing power to the masses and the applications are virtually endless. The Epiphany architecture is made up of a scalable array of independent RISC processors that can all run independent programs simultaneously. This is in great contrast to SIMD based architectures, which can only run one task at any one time. The Epiphany architecture is also unique among massive parallel mobile architectures to allow programming entirely in C/C++. For the developer, this ultimately means faster time to market and the possibility to effectively develop a wide range of high performance applications.

Future Proof Performance Scaling

Epiphany’s multicore architecture was built from day-one to scale perfectly to large array sizes and to keep performance without degradation as we move toward finer process geometries. With the Epiphany, users are ensured a continuous performance increase over time as process nodes keep scaling down, something that is not true of SIMD or cache coherent SMP solutions.

About the Epiphany-IV Multicore Architecture

The Epiphany-IV microprocessor IP, available in Global Foundries’ 28 nanometer low-power processor, operates at up to 800MHz. The Epiphany architecture incorporates up to 4096 cores on a single chip, connected through Adapteva’s patent-pending low-power Network-On-Chip.

Features of the Epiphany architecture include:

• Complete multicore solution featuring a high performance microprocessor ISA, Network-On-Chip, and distributed memory system for seamless integration;
• Fully-featured ANSI C/C++ programmable GNU/Eclipse based tool chain;
• High Performance Superscalar RISC processor cores;
• IEEE Floating Point Instruction Set;
• Shared memory architecture with low-latency local memory at each processor node;
• 25 GB/sec local memory bandwidth;
• 6.4 GB/sec per processor network bandwidth;
• 70 GFLOPS/Watt energy efficiency;
• Processor tile size of 0.13mm2 (with 32KB of SRAM); and
• Processor tile max power of 25mW.

Target Markets and Availability

The Epiphany-IV multicore IP is the ideal technology for enabling the next level of performance in a wide range of battery constrained mobile markets including, machine vision, speech recognition, software defined radio, radar, security and medical diagnostics. Evaluation kits based on an Epiphany-IV 64-core silicon reference design will be available for sampling in Q1 2012, with specialized versions planned for the following quarters. Customers who want immediate platform access are referred to the Epiphany-III 16-core 65nm silicon reference platform, currently in release mode.