The Larrabee Texture Sampler

One of the fixed-function units in the Larrabee processor is a texture sampler. Like the texture unit in a graphics processor, it is capable of performing the usual texture operations, like texture decompression and anisotropic filtering. It communicates with the x86 processing cores via the L2 cache so data access is even faster.

Intel chose to use a fixed-function texture sampler because texture filtering requires specialized data access to unaligned 2x2 blocks of pixels. Texture filtering is also optimized for 8-bit colour values. Finally, the dedicated texture sampler can perform texture filtering 12X faster (40X faster if texture decompression is necessary) than the x86 core.

How Does It Differ From A GPU?

Although the Larrabee processor has a few key components found only in traditional GPUs, it is still a CPU at heart. According to Intel, these key features differentiate the Larrabee from typical GPUs :

1. Each Larrabee core is a complete x86 core that supports :

• context switching and pre-emptive multi-tasking
• virtual memory and page swapping
• fully-coherent caches at all levels

2. Efficient inter-block communication

• Ring bus for full inter-processor communication
• Low-latency high-bandwidth L1 and L2 caches
• Fast synchronization between cores and caches

3. The fixed function logic does not get in the way

• No backend blender between cores and memory
• No rasterization logic between the vertex and pixel stages
• This allows for flexible load balancing and better general functionality

The flexibility of the Larrabee architecture, in Intel's opinion, allows it to be more efficient than a traditional GPU. This is because there is a large variance in the workload in every game. Here's a chart provided by Intel that demonstrates the degree of variance in the workload of these three games.

Support Tech ARP!

If you like our work, you can help support out work by visiting our sponsors, participate in the Tech ARP Forums, or even donate to our fund. Any help you can render is greatly appreciated!