Multi gpu DML?

[TashaSkyUp]

I'm considering getting another gpu first i need to know if tensorflow-directml will use both gpus when training one model?

[adtsai]

Hi, sorry for the delay in response. Multi-GPU distributed training is on our backlog for the coming year, but it isn't something that's super well supported at the moment. Though if you're considering a specific multi-GPU configuration, we'd be interested to hear what you have in mind!

[TashaSkyUp]

Probably an AMD 5700x + 6900x(when it comes out), but if multi gpu training is not coming out until maybe next year.. I'll be forced to buy an nvidia card. And at that point I'll just use regular tensorflow 2.0+ . Sick of Nvidia's stranghold on the market. I'm glad you all re doing this project but ugh... You should get together with AMD and make some sort of arrangement..

[TashaSkyUp]

Honestly its actually the responsibility of google/tensorflow... maybe with this coming lawsuit they will take a look at what other shady anti-competitive things they are doing... Bah maybe i should just swich to pytorch... but i need C# integration.. double bah!

[wchao1115]

@TashaSkyUp Thanks for your interest. We do share the same interest in seeing multi-GPU support become real for this project. This is a matter of how fast we can churn through our current backlog on the more pressing issues e.g. kernel coverage and stability. We do hope that by the time we get to look closely into supporting this, there will be someone like yourself whom we can bounce our ideas with. So stick with us, we're getting there soon.