Larrabee (microarchitecture)

Lore & Background

Larrabee was a hybrid between a multi-core CPU and a GPU, with a coherent cache hierarchy and x86 architecture compatibility like a CPU, and wide SIMD vector units and texture sampling hardware like a GPU. Its x86 cores were based on the simpler P54C Pentium design, superscalar but without out-of-order execution, similar to the Bonnell microarchitecture used in Atom. Each core contained a 512-bit vector processing unit able to process 16 single precision floating point numbers at a time, with scatter/gather instructions and a mask register. The chip included one major fixed-function graphics hardware feature: texture sampling units for trilinear and anisotropic filtering and texture decompression. It had a 1024-bit ring bus for communication between cores and memory, configurable for products with 16 cores or more, or fewer than 16 cores. Each core supported four-way interleaved multithreading.

On December 4, 2009, Intel officially announced that the first-generation Larrabee would not be released as a consumer GPU product, but instead as a development platform for graphics and high-performance computing. On May 25, 2010, the Technology@Intel blog announced Larrabee would be released as a product for high-performance computing competing with the Nvidia Tesla. The Intel MIC multiprocessor architecture announced in 2010 inherited many design elements from Larrabee but does not function as a GPU; the prototype card was named Knights Ferry, with a production card named Knights Corner planned for 2012 or later.

Larrabee drew criticism from GPU competitors. At NVISION 08, an Nvidia employee called Intel's SIGGRAPH paper about Larrabee 'marketing puff' and quoted an industry analyst who speculated the architecture was 'like a GPU from 2006'. By June 2009, Intel claimed prototypes were on par with the Nvidia GeForce GTX 285. At Supercomputing 2009, Intel CTO Justin Rattner demonstrated an overclocked Larrabee processor topping one teraFLOPS, claiming the first public demonstration of a single-chip system exceeding one teraFLOPS, but noted this was early silicon. Because this was only one fifth that of available competing graphics boards, Larrabee was cancelled as a standalone discrete graphics product on December 4, 2009.

Reader's Guide

Larrabee's significance lies in its attempt to create a fully programmable graphics pipeline using many simple x86 cores, in contrast to contemporary GPUs which were only partially programmable. It was intended to support traditional rasterized 3D graphics (Direct3D and OpenGL) for games, while also being suitable for general purpose GPU tasks like ray tracing or physics processing. Its philosophy of using many small, simple cores was similar to the Cell processor, but with key differences: all Larrabee cores were identical, and it used an automatically managed coherent cache hierarchy rather than explicit DMA operations. The project's cancellation in May 2010 led to its technology being passed on to the Xeon Phi line of high-performance computing coprocessors. Almost a decade later, on June 12, 2018, the idea of an Intel dedicated GPU was revived, eventually becoming the Intel Xe and Intel Arc series released in September 2020 and March 2022 respectively, though these were unconnected to the work on the Larrabee project. Larrabee's legacy is thus as a research project that influenced Intel's HPC direction but did not directly lead to a consumer GPU product.

Did You Know?

• The project was terminated in May 2010 and its technology passed on to the Xeon Phi.
• Larrabee's x86 cores were based on the P54C Pentium design, similar to the Bonnell microarchitecture used in Atom.