TeraScale
ATI and AMD graphics microarchitecture using VLIW design
TeraScale is a family of graphics processing unit microarchitectures developed by ATI Technologies and later AMD. It introduced unified shaders and very long instruction word designs to replace fixed pipelines. The architecture powered Radeon HD series cards and early accelerated processing units from 2007 until it was replaced by Graphics Core Next.
- Introduced unified shader model to replace fixed pipelines
- Used very long instruction word design for parallel processing
- Powered Radeon HD 2000 to HD 6000 series graphics cards
- Added hardware tessellation support in later versions
- Replaced by Graphics Core Next architecture in 2011
What it is
TeraScale is a microarchitecture for graphics processing units created by ATI Technologies and continued by AMD. It marked a shift from fixed-function pipelines to a unified shader model, allowing flexible processing of vertex, geometry, and pixel tasks. This design used very long instruction word technology to execute operations in parallel. It served as the foundation for several generations of Radeon graphics cards and integrated graphics in accelerated processing units.
How it came about
The architecture debuted in 2007 with the Radeon HD 2000 series. It competed directly with Nvidia’s Tesla architecture. AMD detailed the design at technical conferences like SIGGRAPH in 2008. The family evolved through three main versions. TeraScale 1 launched with the R600 core. TeraScale 2 arrived in 2009 with the Evergreen generation. TeraScale 3 followed in 2010 with the Northern Islands series. AMD eventually replaced this lineage with the Graphics Core Next architecture in 2011.
Inside the chip
The core uses a very long instruction word single instruction multiple data approach. Shader clusters contain five stream processing units that work in parallel. Four units handle basic math while the fifth manages complex functions like sine. An ultra-threaded dispatch processor manages many threads to keep the hardware busy. Later versions added hardware tessellation units to subdivide polygon meshes. TeraScale 3 reduced the cluster size to four units to improve efficiency. The design relies heavily on compiler optimization to schedule instructions effectively.
How it was made
Manufacturing processes varied across the product life. Early chips like the Radeon HD 2000 used 80 nm and 65 nm nodes. The HD 3000 and 4000 series moved to 65 nm and 55 nm. The HD 5000 and 6000 series utilized 40 nm technology. Die shrinks were common for mid-cycle updates. The architecture supported various memory bus widths depending on the specific chip model.
Where it went
This microarchitecture powered the Radeon HD 2000 through HD 6000 series graphics cards. It also appeared in AMD accelerated processing units codenamed Brazos, Llano, Trinity, and Richland. These chips supported gaming, high-definition video decoding, and multi-monitor setups via Eyefinity technology. The hardware enabled DirectX 11 features in later iterations, improving visual fidelity in games and professional applications.
Versions and family
TeraScale 1 used a five-way very long instruction word design. TeraScale 2 kept this structure but added DirectX 11 support and double precision math. TeraScale 3 shifted to a four-way design to boost performance per clock. Specific chip codenames included Cedar, Cypress, Juniper, and Cayman. Each variant adjusted the number of shader clusters and memory interfaces to target different market segments.
Why it matters
TeraScale established the unified shader model for AMD graphics. It brought hardware tessellation to mainstream PCs before it became mandatory in DirectX 11. The architecture allowed AMD to compete in the discrete graphics market for several years. Its reliance on compiler efficiency influenced driver development strategies. The transition to Graphics Core Next later addressed some limitations of the very long instruction word approach.
Problems and reception
Developers noted that performance depended heavily on instruction mix and compiler quality. The VLIW design faced challenges when instructions had dependencies. Custom filter anti-aliasing offered programmability but could impact speed. Support for open-source drivers was discussed in community forums. The tessellation implementation in early versions used vendor-specific extensions before standardization.
Questions people ask about TeraScale
What does VLIW mean in TeraScale?
Very long instruction word allows the processor to execute multiple operations in a single clock cycle. TeraScale groups five processing units together to handle parallel tasks efficiently.
Which graphics cards used TeraScale?
It powered the Radeon HD 2000, 3000, 4000, 5000, and 6000 series. It was also found in AMD accelerated processing units like Llano and Trinity.
Did TeraScale support DirectX 11?
TeraScale 1 did not fully conform. TeraScale 2 and 3 added full support for DirectX 11 and OpenGL 4.0, including hardware tessellation capabilities.
Who developed the TeraScale architecture?
ATI Technologies originally designed the microarchitecture. AMD continued its development after acquiring ATI, releasing several generations of chips based on this design.
Good reading and sources
The facts on this page come from the Wikipedia article "TeraScale (microarchitecture)", 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.
Like TeraScale
All graphics processorsMore from ATI Technologies
All of them
🇺🇸 United StatesPhenom
AMD desktop processors based on the K10 microarchitectureCPU · x86 · 2007