Are there plans to work on multi-GPU performance with Vulkan?

[digitalscream]

As it says, really - I've got a pair of R9700s running under Vulkan, and currently the only viable options are -sm layer, -sm row (which falls back to layer) and -sm tensor (with a 70% performance penalty).

Does anyone know if there any plans to optimise the row or tensor split strategies for Vulkan?

EDIT: On re-reading, I realise this might come off a bit demanding or critical. That's not the intention at all - it's a genuine question, rather than stamping my foot and insisting people do free work on my behalf ;) I'd chip in myself, but I'm no C++ dev.

[AbdullahMPrograms]

Following as well!

Tensor Parallel looks fantastic on CUDA, however for us multi AMD GPU folk we are unfortunately still limited to Layer split.

[MT5AD]

I know it's a largely meaningless gesture, but I'd be interested as well.