Hunk'a Memory Video Card?

Discussion in 'Videocards - NVIDIA GeForce' started by Zloth, Aug 31, 2015.

  1. Zloth

    Zloth Member Guru

    Messages:
    141
    Likes Received:
    0
    GPU:
    eVGA Nvidia GTX 780
    As I understand it, DX12 is supposed to allow sharing of video card resources. Does that mean that you could buy and drop in, say, an old 700-series card with 2GB of DDR5 and get a video memory boost for cheap? Or would it be better just to get more main memory and let video memory spill over into that?

    Obviously it would only help with DX12 games specifically designed to take advantage of the feature. It will probably be some time before one of those exists. Still, I'm curious how this memory sharing thing is supposed to work and what cards are likely to work better than others.
     
  2. theoneofgod

    theoneofgod Ancient Guru

    Messages:
    4,664
    Likes Received:
    283
    GPU:
    RX 580 8GB
    We don't know if developers will make use of this and if they do it might only be supported in a few games. It's most likely limited by PCI-E bandwidth too so I wouldn't get excited about it until we get some proof of concepts.
     
  3. yasamoka

    yasamoka Ancient Guru

    Messages:
    4,869
    Likes Received:
    255
    GPU:
    EVGA GTX 1080Ti SC
    What it means is that VRAM content won't be mirrored anymore, for games that implement this feature.

    If the GPU in question is being accessed, then it will have its own VRAM content which would be different from another (e.g. main) GPU.

    However, if some resources have to stay in both pools of VRAM (e.g. textures), then we cannot simply sum the VRAM capacity on multiple cards and get a total usable value.

    As for the cards that work together using this feature, we do not know yet. There may be restrictions based on brand (Nvidia, AMD), architecture (Kepler, Maxwell, GCN), family (GTX900 series, Rx-200 series), etc... The feature was demonstrated, though, in a particular DX12 demo, on an Nvidia GPU + Intel iGPU combination.

    One thing is for sure, you can't have less VRAM than if the contents were mirrored so this is likely to bring a boost to total VRAM available unless the main GPU has to hold all content in VRAM at which point you're back to the limitations of a single GPU's VRAM capacity.
     
  4. Zloth

    Zloth Member Guru

    Messages:
    141
    Likes Received:
    0
    GPU:
    eVGA Nvidia GTX 780
    OK, so the jury is still out and probably will remain so until developers get a lot more time in the trenches with DX12. I was hoping we would know better once the documentation got finalized and people started playing around with DX12 demos.

    Thanks for the info!
     

Share This Page