Integrated GPU memory on Metal for Mac

Discussion in 'Mac Programming' started by asiga, Jun 10, 2015.

  1. asiga macrumors 6502a

    Joined:
    Nov 4, 2012
    #1
    One of the points I dislike most from OpenGL for Mac is the limited system memory allocation on integrated GPUs. I find it absurd that you get a 16GB system with integrated GPU and that you cannot use 10GB for textures on massive imaging apps, for example. The OpenGL for Mac implementation neglects that you might wish to use just 4GB RAM for OS X, and the rest 12 GB for graphics (call it unbalanced, but some applications like medical visualization or geography data need such "unbalancing").

    Many years ago, Silicon Graphics released the O2, introducing UMA ("Unified Memory Architecture"), which allowed you to use the whole system RAM for either the CPU or graphics. It's similar to the integrated GPU concept, and in fact I was able to allocate almost 90% of the system RAM for textures on one app I wrote.

    I'm wondering how memory management will be on Metal for Mac, on integrated GPUs. One of the shiny points in Metal is that both the CPU and the GPU can access the same memory buffers (on iOS there's no discrete GPU, so memory is always shared, and it's straightforward to share memory objects between the CPU and the GPU).

    Is Apple saying any details about this on this WWDC?

    Do you know if Metal will be able to use most of the system RAM for graphics, without the current limitation of integrated GPUs on OpenGL?

    Thanks!
     
  2. PBG4 Dude, Jun 12, 2015
    Last edited: Jun 12, 2015

    PBG4 Dude macrumors 68020

    PBG4 Dude

    Joined:
    Jul 6, 2007
    #2
    You want to visit this link --> https://developer.apple.com/metal/
    All the Frameworks and Reference Guides are available without even logging in. "The Metal Programming Guide" is where I think you want to start.

    I don't know much (barely anything) about 3D graphics, but considering there is a paragraph titled "Maintaining Coherency Between CPU and GPU Memory", I'm guessing you may be disappointed with your findings.

    From the video I watched, it seems Metal's focus is freeing up the CPU so it can do more calculations per cycle. They said the reduction was the removal of overhead caused by OpenGL operations, but as I mentioned earlier, I'm the dunce in the corner when it comes to 3D graphics programming.

    If you have an Apple TV, all of WWDC's sessions have been posted, including "What's new in Metal" parts 1 and 2. At 40 minutes each, it may be worth a viewing.

    EDIT: These videos are available from Apple's Developer website.
     

Share This Page