Intel processor graphics GPUs have two main groups of hardware:

*Execution Units/EUs: General purpose execution units. These are used for graphics rendering, but they are also suited to a wide range of media processing tasks. * Media Fixed Function: In addition, specialized fixed function hardware accelerates video codec and frame processing algorithms for fundamentally higher performance at lower power than the EUs or CPUs.

The basic unit of the GPU architecture is the subslice. This contains EUs and memory components. They are assembled into slices.

Subslices are assembled into slices

Execution Units are general purpose compute engines ideal for video processing uses. They are often used by encode for parts of the algorithm not run in fixed function like rate control and mode decisions. The samplers are also highly important to media tasks. They are used by resize and motion estimation.

In addition to EU slices, there is an “unslice” with additional hardware engines individually schedulable for media tasks:

1. VDBox (also known as MFX) for codec operations
2. VEBox (also known as VQE) provides hardware acceleration for video enhancement/frame processing operations.

VDbox/MFX contains:

• Bitstream decoder (BSD).
• ENC (intra prediction, motion estimation)
• PAK (quantization, entropy coding, pixel reconstruction, motion compensation)

VEbox/VQE contains:

• Denoise
• Advanced Deinterlace (ADI)
• Local Adaptive Contrast Enhancement (LACE)
• Camera processing features (skin tone enhancement, etc.)

Intel FP32 GPU performance is still very low

Info from https://software.intel.com/sites/default/files/managed/09/02/Intel_Media_Developers_Guide.pdf