Difference between revisions of "ORCA SOM/ORCA Hardware/Peripherals/NPU"

From DAVE Developer's Wiki
Jump to: navigation, search
 
(One intermediate revision by the same user not shown)
Line 3: Line 3:
 
! colspan="4" style="width:100%; text-align:left" ; border-bottom:solid 2px #ededed" |History
 
! colspan="4" style="width:100%; text-align:left" ; border-bottom:solid 2px #ededed" |History
 
|-  
 
|-  
! style="border-left:solid 2px #73B2C7; border-right:solid 2px #73B2C7;border-top:solid 2px #73B2C7; border-bottom:solid 2px #73B2C7; background-color:#73B2C7; padding:5px; color:white" |Version
 
 
! style="border-left:solid 2px #73B2C7; border-right:solid 2px #73B2C7;border-top:solid 2px #73B2C7; border-bottom:solid 2px #73B2C7; background-color:#73B2C7; padding:5px; color:white" |Issue Date
 
! style="border-left:solid 2px #73B2C7; border-right:solid 2px #73B2C7;border-top:solid 2px #73B2C7; border-bottom:solid 2px #73B2C7; background-color:#73B2C7; padding:5px; color:white" |Issue Date
 
! style="border-left:solid 2px #73B2C7; border-right:solid 2px #73B2C7;border-top:solid 2px #73B2C7; border-bottom:solid 2px #73B2C7; background-color:#73B2C7; padding:5px; color:white" |Notes
 
! style="border-left:solid 2px #73B2C7; border-right:solid 2px #73B2C7;border-top:solid 2px #73B2C7; border-bottom:solid 2px #73B2C7; background-color:#73B2C7; padding:5px; color:white" |Notes
 
|-
 
|-
| style="border-left:solid 2px #73B2C7; border-right:solid 2px #73B2C7;border-top:solid 2px #73B2C7; border-bottom:solid 2px #73B2C7; background-color:#edf8fb; padding:5px; color:#000000" |1.0.0
+
| style="border-left:solid 2px #73B2C7; border-right:solid 2px #73B2C7;border-top:solid 2px #73B2C7; border-bottom:solid 2px #73B2C7; background-color:#edf8fb; padding:5px; color:#000000" |2021/02/04
| style="border-left:solid 2px #73B2C7; border-right:solid 2px #73B2C7;border-top:solid 2px #73B2C7; border-bottom:solid 2px #73B2C7; background-color:#edf8fb; padding:5px; color:#000000" |Feb 2021
 
 
| style="border-left:solid 2px #73B2C7; border-right:solid 2px #73B2C7;border-top:solid 2px #73B2C7; border-bottom:solid 2px #73B2C7; background-color:#edf8fb; padding:5px; color:#000000" |First release
 
| style="border-left:solid 2px #73B2C7; border-right:solid 2px #73B2C7;border-top:solid 2px #73B2C7; border-bottom:solid 2px #73B2C7; background-color:#edf8fb; padding:5px; color:#000000" |First release
 
|-
 
|-
Line 14: Line 12:
 
|}
 
|}
 
<section end="History" />
 
<section end="History" />
<section begin="Preliminary" />
 
  
 
==Peripheral NPU ==
 
==Peripheral NPU ==
 
{{Wip|text=Documentation under NXP's NDA: please refer to helpdesk@dave.eu }}
 
 
<section end="Preliminary" />
 
 
<!--
 
  
 
<section begin="Body" />
 
<section begin="Body" />
Line 43: Line 34:
 
* Data transfers between Neural Network Engines and the Parallel Processing Unit, with SRAM as local storage
 
* Data transfers between Neural Network Engines and the Parallel Processing Unit, with SRAM as local storage
 
* Neural Network Engine and Parallel Processing Unit synchronization with hardware semaphore
 
* Neural Network Engine and Parallel Processing Unit synchronization with hardware semaphore
 
-->
 
  
 
----
 
----
  
 
[[Category:ORCA]]
 
[[Category:ORCA]]

Latest revision as of 18:39, 28 December 2023

History
Issue Date Notes
2021/02/04 First release


Peripheral NPU[edit | edit source]


The Neural Processign Unit (NPU) available on ORCA is based on iMX8MPlus SoC.

Description[edit | edit source]

The Neural Processing Unit (NPU) core accelerates vision image processing functions and provides enhanced performance for real-time use cases with hardware support for the OpenVX API.

Key features of the NPU block include:

  • OpenVX 1.2 compliance, including extensions
  • Convolutional Neural Network acceleration
  • IEEE 32-bit floating-point pipeline in PPU shaders.
  • Ultra-threaded parallel processing unit
  • Low bandwidth at both high and low data rates
  • Low CPU loading
  • MMU functionality supported
  • Performance Counters for DMA Profiling
  • Data transfers between Neural Network Engines and the Parallel Processing Unit, with SRAM as local storage
  • Neural Network Engine and Parallel Processing Unit synchronization with hardware semaphore