project D415 hardware requirements

• MartyG

  • April 07, 2022 22:02

  Hi Aljosa  Any of the listed i7 Intel Core CPUs that are modern 11th Generation (indicated by the '11' in their CPU model code) should be sufficient. An i5 should be able to handle two cameras simultaneously, but if your project budget allows then an i7 would be better.

   

  RealSense 400 Series cameras have built-in hardware called the Vision Processor D4 that can perform some of the camera data processing that the computer CPU / GPU would otherwise have to do.  This enables the cameras to be used successfully with lower-specification computers. 

  However, there are some processing operations (such as depth-RGB alignment, point clouds and post-processing filters) that will be processed on the computer hardware rather than the camera, so a good computer specification will be advantageous in that situation.  Skeleton tracking can also be a processing-intensive application and so could benefit from a strong hardware specification.

   

  A computer with a good in-built GPU will be an advantage (not necessarily the latest top-end gaming GPUs).  Applications created in the RealSense SDK in the C++ language can take advantage of a system called GLSL Processing Blocks to offload processing from the CPU onto the GPU to accelerate SDK functions such as alignment and point clouds that belong to the rs2:: class.  Improvement from use of GLSL will not be noticeable on low-end computing devices.  GLSL is 'vendor neutral' and so can work with any GPU brand.

  Alternatively, if your computer has an Nvidia GPU chip or video card then support for CUDA can instead be enabled in the SDK to accelerate color conversion, alignment and point cloud operations.  C++ is not necessary for this as CUDA will be enabled automatically if building the SDK from Debian packages or by building it from source code with CMake and including a particular CUDA-enabling term in the CMake build command.

  I would strongly recommend selecting a computer that has some form of built-in GPU at a minimum rather than no GPU at all.

   

  4 cores (quad core) should be sufficient as a minimum for a processing-intensive application, especially as CPU work can be offloaded to a GPU to reduce CPU percentage usage.  The SDK can be built though with support for making use of multiple cores when performing YUY to RGB color conversion and depth-RGB alignment, at the expense of greater CPU utilization.

   

  Intel Iris XE is considered to be a good gaming-capable GPU.  So it should be suitable for use with RealSense cameras.  

   

  If your project is able to make use of a desktop over a laptop (for example, portability and space-saving is not a concern) then a desktop would be preferable in my opinion.

  1

  Comment actions Permalink
• MartyG

  • April 07, 2022 22:11

  Also, if you would prefer a small form-factor that lends itself to portability but need desktop power, an Intel NUC may suit you.  It is available in a wide range of price-points and configurations from modest to highly powerful, and in board, kit or complete ready-to-run product formats that are very space-efficient.

  https://www.intel.co.uk/content/www/uk/en/products/details/nuc.html

  1

  Comment actions Permalink
• Aljosa

  • April 08, 2022 07:01

  Thank you very much MartyG! It's clear now. Best, alex

  0

  Comment actions Permalink

Please sign in to leave a comment.