Tesla (microarchitecture)

Tesla (microarchitecture) was a GPU microarchitecture developed by Nvidia and introduced in 2006. It represented a fundamental shift from the traditional fixed-function pipeline to a unified shader architecture, enabling more general-purpose computing. The architecture formed the foundation for the GeForce 8 series and subsequent consumer graphics cards, as well as the company's first dedicated GPGPU platform, Tesla.

Overview

The Tesla architecture, first implemented in the G80 GPU, marked a revolutionary departure from its Curie predecessor by introducing a fully unified shader model. This design allowed a single array of stream processors to handle vertex, geometry, and pixel shading tasks dynamically, greatly improving efficiency. The architecture was central to Nvidia's strategy for the DirectX 10 era, enabling support for Shader Model 4.0 and new features like geometry shaders. Its introduction coincided with the launch of the GeForce 8800 GTX, which set new performance standards for consumer graphics.

Architecture

At its core, the Tesla architecture organized its stream processors into groups called Streaming Multiprocessors (SMs), each containing scalar processors for arithmetic logic and special function units. It featured a unified L1 cache and shared L2 cache to manage memory traffic between the shader cores and the video memory. Key innovations included the decoupling of shader clock speeds from the core clock and the introduction of a hardware thread scheduler to maximize processor utilization. Later refinements, such as the GT200 GPU, increased the number of processors and added support for IEEE 754-2008 floating-point precision and Shader Model 4.1.

Products

Tesla-powered consumer products spanned multiple generations of Nvidia's GeForce brand, starting with the GeForce 8 series (G80/G92) and including the GeForce 9 series, GeForce 100 series, GeForce 200 series, and GeForce 300 series. In the professional and high-performance computing space, the architecture was used in the Quadro FX and Quadro VX lines, as well as the initial Nvidia Tesla brand of GPGPU computing modules like the Tesla C870. It also appeared in ION platform chipsets and certain GeForce 8M series for laptops.

Performance and reception

Upon release, Tesla-based cards like the GeForce 8800 GTX delivered unprecedented performance in contemporary titles such as Crysis and Call of Duty 4: Modern Warfare, dominating benchmarks from sites like AnandTech and Tom's Hardware. The architecture's general-purpose computing capabilities, exposed through CUDA, were leveraged for accelerating tasks in Folding@home and early ray tracing research. However, it faced criticism for high power consumption and thermal output in its flagship designs, and its initial lack of support for DirectX 10.1 was noted by competitors like AMD.

Successor and legacy

Tesla was succeeded in 2010 by the Fermi architecture, which introduced major enhancements like an improved cache hierarchy, support for ECC memory, and more robust GPGPU features. The legacy of Tesla is profound, as its unified shader model became the standard for all subsequent GPU designs from Nvidia and its rivals. It established CUDA as a dominant parallel computing platform, influencing fields from scientific computing to artificial intelligence, and its architectural principles directly informed the development of later microarchitectures like Kepler and Maxwell.

Category:Nvidia microarchitectures Category:Graphics processing units Category:2006 in computing