From DigitalVision
Jump to: navigation, search


GPU Frequently Asked Questions

[ Also see the General FAQ ]

A Note on Previous Usage

We have always used the GPU in the Nucoda and Phoenix products. It was used by the Master Layer and also by the multiple output modes (i.e. the real time scaling of an image).

From release 2017.1 we have extended this and now all colour layers are GPU previewed.

What should I expect to see in this new version?

  • We wanted to make this new version as seamless as possible. To be honest it may look very much the same to you. When working we expect things may be a bit smoother, working in UHD will allow you to do more in real time with your colour tools, adding more layers. But since we’ve always had such effective CPU rendering it may not feel like much has changed.
  • As the CPU will now be used less it is quite probable that when working, you will get better performance from compressed media since there are more processor cycles to use.

Do you use Cuda or OpenCL?

  • We do not use Cuda or OpenCL - our tools are based on GLSL shaders

Which tools work on the GPU?

  • All the colour tools run on the GPU, as well as :
    • Pan & Scan
    • Router
    • Blur (Note: Blur is very GPU intensive especially in UHD)

Which tools are not on the GPU and why?

  • Shapes are currently still created and rendered on the CPU, then uploaded to the GPU as mattes - this will change in a later version.
  • Keys are currently still created and rendered on the CPU, then uploaded to the GPU as mattes - this will change in a later version.
  • Keys can run in real-time depending on the blur type used, (it is possible to use a simple blur and get real time performance, however, using the despeckle (median) filter will force the key to be rendered.
  • Dissolves and Dynamics will still be CPU rendered.
  • Interlaced material will use the GPU but blurs and Pan & Scan will cause a CPU cache to be required.
  • DVO Tools are CPU only
  • OFX Plugins like Tiffen Filters or Genarts Sapphire - the Sapphire tools can make use GPU to accelerate processing. The tools are limited to using a single GPU and although some can play back fast enough for previews, they will all require caching. Performance in UHD will be be slower. Use Interactive mode to increase interactivity and previews. This is no different to any other supported host.

Why release a version where everything isn’t on the GPU?

  • We feel that there is a significant benefit to having the colour layers on the GPU and releasing the support in stages will allow this benefit to get to customers quicker.

What about Floating Point grading?

  • Although the interactive part of the grading is on the GPU, the actual frames for caching and output is still based on the CPU, for this reason, if you want to have the benefit of working in floating point you will need to work in a Half project, with material set to half or use the Float processing project option.
  • You can also set material to half only when needed, this will save processing time and space.
  • When working with 10 bit media Nucoda will behave as it always have, clipping between layers. Setting the media to Half will prevent clipping between layers but caches will be OpenEXR files which are larger and will impact performance
  • When working in Half and writing EXR files, blurs are especially CPU heavy and will impact performance.

Do I still have to cache media?

  • Caching is still very important for Nucoda and Phoenix and will always be an important part of ensuring reliable playback. When using any tools in the Input FX layer, the material will still be cached to create new source material.
  • For Nucoda we have made the easy decision to keep the caching - it has been out there the longest and is the most robust of any of the caching systems available today. It lets us always have a version we can play back - whatever we add to the grading stack.
  • We have improved our Interactive Mode tool - users can preview and run in a reduced resolution on the fly, it allows many Sapphire effects to be played back in real time or close to it for preview.
  • The GPU version adds timing controls, which are experimental, the idea is that the system will time what the user is doing and build a profile of the time it takes to do certain operations.
    • So while you are using the colour tools we will let you add a lot of layers, even with shapes and blurs. As soon as the systems recognises that it might not be able to play in real time, it will automatically add a cache and render at that layer, and then let the user keep working using the GPU.
  • When another effect needs to be rendered, the cache will simply move to the new layer and the old one is deleted. Inside the router there is a button for the user to force a render if preferred.

Do you support multiple GPU’s?

  • Currently we only support a single GPU (this might change), but the recommended GPU’s are very powerful and will allow you to work in UHD.

Which GPU’s do you support?

  • We do QA and officially support Nvidia M5000, M6000, P5000 and P6000.
    • Some users are testing using Titan X and GTX1080ti GPU’s
  • Other GPU’s will most likely work (Nvidia and AMD) but they are not tested. For working in UHD you will require a larger amount of RAM and Gen 3 cards.