5

I've been looking into using vulkan to leverage compute capabilities on integrated graphics and discrete graphics capabilities. However, according to this article, despite having multi gpu support through compatible SLI configurations, Vulkan does not have heterogeneous multi gpu support. The same article goes on to claim the DX12 does in fact have this kind of support, going "a step further".

In my application I want to take advantage of the compute capabilities of two separate devices which do not have SLI/Xfire compatibility, but are also not necessarily working on the same compute workload, memory need not be shared between each to compute their workloads.

Is it not possible for me to simply have two devices do two separate things in the same instance, and if necessary, share memory via copying to host first? This answer appears to imply that I can actually do that, so I'm confused as to what heterogeneous multi gpu actually means for DX12, and how that is different from what I can do in Vulkan.

Krupip
  • 4,404
  • 2
  • 32
  • 54
  • 1
    You can create several logical devices in a single Vulkan instance. And each of these devices can perform its own calculations and/or rendering. No heterogeneous multi-GPU support means You cannot share resources between different devices, like for example discrete and integrated GPU. But nothing stops You from using them individually at the same time, but each of them as a separate entry/device. – Ekzuzy May 15 '18 at 19:34
  • @Ekzuzy That is great info, thanks, I'm still looking for what the heck DX12 is doing, because I don't see how DX12 can have heterogeneous multigpu support, the only way you'd be able to share information is through device->host->device, which is exactly what you'd be able to do in vulkan... – Krupip May 15 '18 at 19:47
  • I don't know DX12 at all so I cannot compare it with Vulkan. But maybe it's a good marketing? ;-) But copying data to the CPU and then back again to another logical device shouldn't be perceived as "resource sharing". – Ekzuzy May 15 '18 at 20:06
  • You can look at [Multi-Adapter MSDN article](https://msdn.microsoft.com/en-us/library/windows/desktop/dn933253(v=vs.85).aspx). – user7860670 May 15 '18 at 21:11
  • @VTT Thanks, I've already looked at that, and unfortunately it didn't answer my question. I've not seen MS be specific at all what it means to have "heterogeneous multi GPU support" – Krupip May 16 '18 at 13:17
  • 1
    Should we be interpreting another person's words? Why not ask the author of the original article directly? – krOoze May 16 '18 at 20:30

0 Answers0