Cycles:

* OpenCL now only uses GPU/Accelerator devices, it's only confusing if CPU
  device is used, easy to enable in the code for debugging.
* OpenCL kernel binaries are now cached for faster startup after the first
  time compiling.
* CUDA kernels can now be compiled and cached at runtime if the CUDA toolkit
  is installed. This means that even if the build does not have CUDA enabled,
  it's still possible to use it as long as you install the toolkit.

intern/cycles/util/util_cuda.cpp

@@ -16,6 +16,8 @@
  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
  */
 
+#include <stdlib.h>
+
 #include "util_cuda.h"
 #include "util_debug.h"
 #include "util_dynlib.h"
@@ -371,6 +373,11 @@ bool cuLibraryInit()
 	/* cuda 4.0 */
 	CUDA_LIBRARY_FIND(cuCtxSetCurrent);
 
+#ifndef WITH_CUDA_BINARIES
+	if(cuCompilerPath() == "")
+		return false;
+#endif
+
 	/* success */
 	result = true;
 
@@ -379,13 +386,23 @@ bool cuLibraryInit()
 
 string cuCompilerPath()
 {
-		/* todo: better nvcc detection */
 #ifdef _WIN32
-	string nvcc = "C:/CUDA/bin/nvcc.exe";
+	const char *defaultpath = "C:/CUDA/bin";
+	const char *executable = "nvcc.exe";
 #else
-	string nvcc = "/usr/local/cuda/bin/nvcc";
+	const char *defaultpath = "/usr/local/cuda/bin";
+	const char *executable = "nvcc";
 #endif
 
+	const char *binpath = getenv("CUDA_BIN_PATH");
+
+	string nvcc;
+
+	if(binpath)
+		nvcc = path_join(binpath, executable);
+	else
+		nvcc = path_join(defaultpath, executable);
+
 	return (path_exists(nvcc))? nvcc: "";
 }