Keras: Any multi-GPU solution for Theano backend?

[mycrazycracy]

Hey guys,
I used keras 2.0.4 for weeks, and found it really convenient to build my idea. I once use tensorflow backend until I found theano is faster than tensorflow on my server. I double checked my installation and I think there is no problem. I also found many people report theano is actually faster in the forum. Anyway, I change my backend to theano now.
Now, the problem is there is NO OUT-OF-BOX SOLUTION for theano backend to use multiple GPUs. As I have 8 K40s in my server, I would definitely like to utilize them by either data or model parallelism.
I searched all the issues here (or google it) and just found a way to use multiple GPUs for tensorflow backend (e.g. https://github.com/fchollet/keras/issues/2436). But almost one year passed, there is still no such solution for theano. I would very very appreciate it if anyone could help me for this issue!

[mycrazycracy]

I would wonder nobody face this problem or just because I miss something?
I think multi-gpu training is critical in the deep learning software society. Isn't it?
Anyone could help?

[edmondja]

Maybe this can help for multi-gpu https://github.com/Microsoft/CNTK/issues/797.

[mycrazycracy]

Still, it seems like there is no clear answer for this problem. Oooops...

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed after 30 days if no further activity occurs, but feel free to re-open a closed issue if needed.

[Golbstein]

4 months passed and no one could help this poor lad? I'm also wondering what's the issue with that slower tensorflow backend and why there's no multi gpu support in keras