Search Unity

  1. Calling all beginners! Join the FPS Beginners Mods Challenge until December 13.
    Dismiss Notice
  2. It's Cyber Week at the Asset Store!
    Dismiss Notice

GPU

Discussion in 'Cinemachine' started by Dan_Miller, Dec 3, 2019 at 1:08 AM.

  1. Dan_Miller

    Dan_Miller

    Joined:
    Sep 4, 2012
    Posts:
    22
    I'm preparing to learn Cinemachine but making sure my hardware is up to the task. Is there any advantage to using two GPUs for rendering? The idea being to render in real-time.
     
  2. Adam_Myhill

    Adam_Myhill

    Unity Technologies

    Joined:
    Dec 22, 2016
    Posts:
    332
    Hi Dan,

    Cinemchine is a camera animation system. It's pretty light, not usually even noticeable cost-wise in most cases. I think you may be concerned with outright rendering performance. In that case yes, a beefy GPU helps a lot, especially if you need to render in realtime VS just using Recorder and outputting files.

    As for dual cards, this is a good start: https://support.unity3d.com/hc/en-u...00.859192238.1575578723-2128062438.1571424202

    The short answer is no, dual cards won't help due to limitations around preserving framebuffers across frames such as with motion blur, TAA, SSReflections, etc., - if you want those kinds of things. From the article:

    In general, if your content uses results from a previous frame, then SLI will not be faster. The best advice would be to make sure to render everything you need in the same frame. To clarify, you should not render some textures in one frame and use them in another frame. Specifically from the Image Effects package, the "motion blur" effect uses a render texture from the previous frame.

    The size/complexity of your scene will make the most difference. Do you have every post processing effect cranked up? Want to do it at 4K? Have 40 million polygons in your scene? Those will be your biggest bottlenecks.

    A geforce 1080 is probably the minimum to get rock solid 30fps our of a decently complex scene.. but again, a lot of variables with the scene complexity..

    So I'd just get the fastest card you can and load the scene up until you approach minimum acceptable performance.

    If it doesn't have to be real-time, then use Recorder and crank everything up to 11