diff --git a/data/get_mnist.sh b/data/get_mnist.sh
index 2a9ad083fb4..c807a7b9439 100755
--- a/data/get_mnist.sh
+++ b/data/get_mnist.sh
@@ -17,7 +17,7 @@ gunzip t10k-labels-idx1-ubyte.gz
 
 echo "Creating leveldb..."
 
-../examples/convert_mnist_data.bin train-images-idx3-ubyte train-labels-idx1-ubyte mnist-train-leveldb
-../examples/convert_mnist_data.bin t10k-images-idx3-ubyte t10k-labels-idx1-ubyte mnist-test-leveldb
+../build/examples/convert_mnist_data.bin train-images-idx3-ubyte train-labels-idx1-ubyte mnist-train-leveldb
+../build/examples/convert_mnist_data.bin t10k-images-idx3-ubyte t10k-labels-idx1-ubyte mnist-test-leveldb
 
 echo "Done."
diff --git a/include/caffe/common.hpp b/include/caffe/common.hpp
index e7c5abe7435..7fd7ea6329c 100644
--- a/include/caffe/common.hpp
+++ b/include/caffe/common.hpp
@@ -49,7 +49,12 @@ using boost::shared_ptr;
 
 
 // We will use 1024 threads per block, which requires cuda sm_2x or above.
-const int CAFFE_CUDA_NUM_THREADS = 1024;
+#if __CUDA_ARCH__ >= 200
+    const int CAFFE_CUDA_NUM_THREADS = 1024;
+#else
+    const int CAFFE_CUDA_NUM_THREADS = 512;
+#endif
+
 
 
 inline int CAFFE_GET_BLOCKS(const int N) {