Avision processing unit (VPU) is (as of 2023) an emerging class ofmicroprocessor; it is a specific type ofAI accelerator, designed toacceleratemachine vision tasks.^[1][2]

Vision processing units are distinct fromgraphics processing units (which are specialised forvideo encoding and decoding) in their suitability for runningmachine vision algorithms such as CNN (convolutional neural networks), SIFT (scale-invariant feature transform) and similar.

They may includedirect interfaces to take data fromcameras (bypassing any off chip buffers), and have a greater emphasis on on-chipdataflow between manyparallel execution units withscratchpad memory, like aspatial architecture or amanycoreDSP. But, like video processing units, they may have a focus onlow precisionfixed point arithmetic forimage processing.

They are distinct fromGPUs, which contain specialised hardware forrasterization andtexture mapping (for3D graphics), and whosememory architecture is optimised for manipulatingbitmap images inoff-chip memory (readingtextures, and modifyingframe buffers, withrandom access patterns). VPUs are optimized for performance per watt, while GPUs mainly focus on absolute performance.

Target markets arerobotics, theinternet of things (IoT), new classes ofdigital cameras forvirtual reality andaugmented reality,smart cameras, and integrating machine vision acceleration intosmartphones and othermobile devices.

• Movidius Myriad X, which is the third-generation vision processing unit in the Myriad VPU line fromIntel Corporation.^[3]
• Movidius Myriad 2, which finds use inGoogle Project Tango,^[4]Google Clips and DJI drones^[5]
• Pixel Visual Core (PVC), which is a fully programmableImage, Vision andAI processor for mobile devices
• Microsoft HoloLens, which includes an accelerator referred to as aholographic processing unit (complementary to its CPU and GPU), aimed at interpreting camera inputs, to accelerate environment tracking and vision for augmented reality applications.^[6]
• Eyeriss, aspatial architecture designed fromMIT intended for runningconvolutional neural networks.^[7]
• NeuFlow, a design byYann LeCun (implemented inFPGA) for acceleratingconvolutions, using a dataflow architecture.
• Mobileye EyeQ, byMobileye
• Programmable Vision Accelerator (PVA), a7-way VLIW Vision Processor designed byNvidia.

Some processors are not described as VPUs, but are equally applicable to machine vision tasks. These may form a broader category ofAI accelerators (to which VPUs may also belong), however as of 2016 there is no consensus on the name:

• IBMTrueNorth, aneuromorphic processor aimed at similar sensor datapattern recognition and intelligence tasks, including video/audio.
• Qualcomm Zeroth Neural processing unit, another entry in the emerging class of sensor/AI oriented chips.^[8]
• All models of IntelMeteor Lake processors have aVersatile Processor Unit (VPU) built-in for acceleratinginference for computer vision and deep learning.^[9]

• Adapteva Epiphany, a manycore processor with similar emphasis on on-chip dataflow, focussed on 32-bit floating point performance
• CELL, a multicore processor with features fairly consistent with vision processing units (SIMD instructions & datatypes suitable for video, and on-chip DMA between scratchpad memories)
• Coprocessor
• Graphics processing unit, also commonly used to run vision algorithms. NVidia's Pascal architecture includes FP16 support, to provide a better precision/cost tradeoff for AI workloads
• MPSoC
• OpenCL
• OpenVX
• Physics processing unit, a past attempt to complement the CPU and GPU with a high throughput accelerator
• Tensor Processing Unit, a chip used internally by Google for accelerating AI calculations

• Eyeriss architecture
• Holographic processing unit
• NeuFlow: A Runtime Reconfigurable Dataflow Processor for VisionArchived 2017-05-05 at theWayback Machine

• Microprocessors
• Neural processing units
• Machine vision

• Articles with short description
• Short description is different from Wikidata
• Webarchive template wayback links