summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorJeff Donahue <jeff.donahue@gmail.com>2014-08-16 23:47:25 -0700
committerJeff Donahue <jeff.donahue@gmail.com>2014-08-17 01:07:17 -0700
commitf135665a205490540721471dc4c5ea54cb3dc22f (patch)
treea39774e2142b4473afbc7fc0f11ce14eb475869b
parentdb211a27fd811f827ffb684f5f4e4d3f8c94a262 (diff)
downloadcaffeonacl-f135665a205490540721471dc4c5ea54cb3dc22f.tar.gz
caffeonacl-f135665a205490540721471dc4c5ea54cb3dc22f.tar.bz2
caffeonacl-f135665a205490540721471dc4c5ea54cb3dc22f.zip
[docs] CMake build steps and Ubuntu 12.04 install instructions
-rw-r--r--docs/installation.md21
1 files changed, 21 insertions, 0 deletions
diff --git a/docs/installation.md b/docs/installation.md
index ff0760dd..c2bd64d8 100644
--- a/docs/installation.md
+++ b/docs/installation.md
@@ -210,6 +210,27 @@ Be sure to set your MATLAB and Python paths in `Makefile.config` first!
Now that you have installed Caffe, check out the [MNIST demo](mnist.html) and the pretrained [ImageNet example](imagenet.html).
+### Compilation using CMake (beta)
+
+In lieu of manually editing `Makefile.config` to tell Caffe where dependencies are located, Caffe also provides a CMake-based build system (currently in "beta").
+It requires CMake version >= 2.8.8.
+The basic installation steps are as follows:
+
+ mkdir build
+ cd build
+ cmake ..
+ make all
+ make runtest
+
+#### Ubuntu 12.04
+
+Note that in Ubuntu 12.04, Aptitude will install version CMake 2.8.7 by default, which is not supported by Caffe's CMake build (requires at least 2.8.8).
+As a workaround, if you are using Ubuntu 12.04 you can try the following steps to install (or upgrade to) CMake 2.8.9:
+
+ sudo add-apt-repository ppa:ubuntu-sdk-team/ppa -y
+ sudo apt-get -y update
+ sudo apt-get install cmake
+
## Hardware Questions
**Laboratory Tested Hardware**: Berkeley Vision runs Caffe with K40s, K20s, and Titans including models at ImageNet/ILSVRC scale. We also run on GTX series cards and GPU-equipped MacBook Pros. We have not encountered any trouble in-house with devices with CUDA capability >= 3.0. All reported hardware issues thus-far have been due to GPU configuration, overheating, and the like.