merge

post/docker/using-gpu.md

@@ -0,0 +1,32 @@
+Using Host's GPU on Docker
+===
+
+
+There are two steps to using GPU on your Docker:
+
+* Install nvidia-docker (for it's volumes)
+* Setup volumes and device for your container
+
+## FIRST STEP: install nvidia docker
+
+Checkout [Nvidia's official report for nvidia-docker](https://github.com/NVIDIA/nvidia-docker) to get the guides about install
+nvidia-docker, and then install it.
+
+After it installed, you need to make sure `nvidia-docker-plugin` is running.
+If not, you need run it on the background.
+
+Then you need to use `docker volumes ls` to checkout whether there are a volume `named nvidia-docker`.
+You also need to run `nvidia-docker run --rm nvidia/cuda nvidia-smi` to checkout whether the container can connect to
+host GPU.
+
+## SECOND STEP: launch your container
+
+> Before beginning with all of this section, it is important that in your image, there should install cuda toolkit or sdk.
+
+If you want to use GPU, the following list is want you need to do:
+
+* Mount volume `nvidia-docker` to your container by using the option `-v` when creating container
+* Map the devices about GPU by using `--device` when creating container
+  > You need to map those whose name has prefix `nvidia`
+* Launch your container
+

post/docker/using-gpu.sum.md

@@ -0,0 +1,3 @@
+## Summary
+
+So in this post, I will tell your about how to use Nvidia's GPU on Docker.