Teraflops Research Chip
80-core research processor by Intel targeting 1 TFLOPS
The Teraflops Research Chip, codenamed Polaris, is an experimental manycore processor developed by Intel. Released in 2007, it features 80 cores and a network-on-chip design to achieve one teraflop of performance under 100 watts.
- Contains 80 cores arranged in a 10x8 mesh network
- Targets 1 TFLOPS performance under 100 W power draw
- Uses 3D-stacked SRAM for 1 TB/s memory bandwidth
- Manufactured on a 65 nm process with 100 million transistors
- Precursor technology to the Xeon Phi product line
What it is
The Teraflops Research Chip, known internally as Polaris, is an experimental central processing unit created by Intel. It serves as a proof of concept for high-performance computing using many small cores rather than a few large ones. The project aimed to show that a modular design could deliver one trillion floating-point operations per second while keeping power consumption below 100 watts. This chip uses a unique network-on-chip layout to connect its 80 processing units efficiently.
How it came about
Intel’s Tera-Scale Computing Research Program led the development of this processor. Sriram R. Vangal served as the technical lead for the project. The team first showed the design at the Intel Developer Forum in late September 2006. They officially announced the chip in February 2007. A functional version was later demonstrated at the IEEE International Solid-State Circuits Conference that same year, where detailed specifications were shared with the engineering community.
Inside the chip
The chip arranges 80 cores in a two-dimensional mesh grid. Each core, or tile, includes a processing engine and a five-port router. These routers allow data to move quickly across the chip with low latency. The architecture uses a 96-bit very long instruction word format, which lets each core handle multiple operations in one cycle. Each tile has two floating-point units capable of two single-precision calculations per clock cycle. This setup allows for high parallel throughput without complex control logic.
How it was made
Intel manufactured the processor using a 65 nanometre complementary metal-oxide-semiconductor process. The die measures 275 square millimetres and holds 100 million transistors. Eight layers of copper interconnects link the components. To boost memory speed, the design stacks static random-access memory modules directly under each tile. This three-dimensional stacking increases bandwidth significantly but limits total capacity to 20 megabytes and adds thermal challenges.
Where it went
This processor was built for research purposes rather than commercial sale. It targeted high-performance computing tasks that benefit from massive parallelism. Intel used it to explore new programming models, such as Ct, to help developers write software for manycore systems. Although Ct did not become widely adopted, the lessons learned from Polaris influenced later products like the Xeon Phi coprocessors used in supercomputers.
Versions and family
Polaris stands as a singular research prototype within Intel’s Tera-Scale programme. It does not have direct consumer variants. However, its architectural concepts paved the way for the Xeon Phi line, which brought manycore designs to the market. The chip succeeded earlier dual-core and quad-core experiments by scaling up to 80 cores on a single piece of silicon.
Why it matters
The project demonstrated that network-on-chip designs could manage communication between dozens of cores effectively. It achieved high energy efficiency, reaching nearly 20 gigaflops per watt in actual tests. The bisection bandwidth of the internal network proved robust at high frequencies. While the specific programming model failed to gain traction, the hardware insights helped shape future manycore accelerators and server processors.
Problems and reception
The chip was well-received in academic and industrial circles for its innovative approach to core connectivity. Critics noted the complexity of programming such a system. The custom instruction set included specific commands for managing packet traffic and power states. Despite the technical success, the lack of a sustainable software ecosystem limited its broader adoption outside of Intel’s own research labs.
Questions people ask about Teraflops Research Chip
What is the main purpose of the Teraflops Research Chip?
It is a research processor designed to demonstrate that a modular manycore architecture can achieve one teraflop of performance while consuming less than 100 watts of power. It explores network-on-chip communication methods.
How many cores does the Polaris chip have?
The chip contains 80 individual processing cores, often referred to as tiles. These are arranged in a two-dimensional mesh grid to facilitate efficient data movement across the processor die.
What manufacturing process was used for this chip?
Intel manufactured the processor using a 65 nanometre CMOS process. The die includes eight layers of copper interconnects and contains a total of 100 million transistors on a 275 square millimetre surface.
Did this chip lead to any commercial products?
Yes, the research and architectural lessons from the Polaris project were later incorporated into Intel’s Xeon Phi line of coprocessors. These commercial chips brought manycore designs to high-performance computing markets.
Good reading and sources
The facts on this page come from the Wikipedia article "Teraflops Research Chip", written by its contributors and used under the CC BY-SA 4.0 licence, and from its Wikidata record (CC0). EveryChip rewrote the text in its own words, with AI assistance, and compares every page with the article to make sure no passage is copied. It may be reused under the same CC BY-SA 4.0 licence with a link back.
It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.