I read this presentation and on 6th slide I found that is not need to use multiple render targets and depth can be taken directly from depth buffer.
From which version of DirectX or OpenGL can be the feature used? How to do that?
As far as rendering the scene goes, it's just like using the ordinary depth buffer. The difference is that here the depth buffer is a texture which you can later sample in the lighting passes.
Depth textures have been available for a long time - since D3D9, or OpenGL 1.4. You can pretty much assume universal support on desktop GPUs today. For mobile platforms, depth textures are an extension in OpenGL ES 2; I'm not sure how widespread the support is. If you just google for the phrase "depth texture" you should be able to find details of how to do it on your chosen platform/API.
Don't know about OpenGL, but for DirectX it can be done in DX10+. See this article.