Specifications for Deep learning Rig

[Ashar410]

Hey Guys! This might be a very simple question.
1) I need to build a rig for deep learning purposes. I want minimum 3 graphic cards on the board. So which motherboard, chipset and CPU combination is most suited to support this setup. 

2) I want to use maximum PCIe Lanes of graphic cards.

3) Is it necessary to use each card at x16 config?

4) Which is more important? Chipset's support for PCIe lanes or CPU's support for PCIe lanes?

[Radium_Angel]

14 minutes ago, Ashar410 said:

Hey Guys! This might be a very simple question.
1) I need to build a rig for deep learning purposes. I want minimum 3 graphic cards on the board. So which motherboard, chipset and CPU combination is most suited to support this setup. 

2) I want to use maximum PCIe Lanes of graphic cards.

3) Is it necessary to use each card at x16 config?

4) Which is more important? Chipset's support for PCIe lanes or CPU's support for PCIe lanes?

The type of Recurrent Neural Networks that I use are all heavily GPU-based, the CPU doesn't matter at all (to a point, can't be running a Celeron 300a amd expect good results)

And the difference between x16 and x8 is very minor, except for some very specific applications.

 

What DL system software are you planning on using? LUA?