diff --git a/CMakeLists.txt b/CMakeLists.txt
index 9a657421626617b8d236bc4c7e46b6b69cedfdf0..514f120daf28a512659a12b6e86d5fe64f6c9c85 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -113,6 +113,7 @@ endif()
 ###Add Libraries
 if (CUDA_FOUND)
 	cuda_add_cufft_to_target(bioEM)
+	target_link_libraries(bioEM ${CUDA_CUDA_LIBRARY})
 endif()
 
 target_link_libraries(bioEM -L${FFTW_LIBDIR} -lfftw3 -lfftw3f)
diff --git a/bioem_cuda.cu b/bioem_cuda.cu
index e412dfd52631f276017c36bb06b97269cfe8f1d6..18d4369d7abbd9f079fee04d554b39d38a162646 100644
--- a/bioem_cuda.cu
+++ b/bioem_cuda.cu
@@ -272,9 +272,78 @@ int bioem_cuda::compareRefMaps(int iOrient, int iConv, const myfloat_t* conv_map
 	return(0);
 }
 
+int bioem_cuda::selectCudaDevice()
+{
+	int count;
+	
+	long long int bestDeviceSpeed = -1;
+	int bestDevice;
+	cudaDeviceProp deviceProp;
+	
+	checkCudaErrors(cudaGetDeviceCount(&count));
+	if (count == 0)
+	{
+		printf("No CUDA device detected\n");
+		return(1);
+	}
+	for (int i = 0;i < count;i++)
+	{
+		printf("CUDA device %d\n", i);
+#if CUDA_VERSION > 3010
+		size_t free, total;
+#else
+		unsigned int free, total;
+#endif
+		cuInit(0);
+		CUdevice tmpDevice;
+		cuDeviceGet(&tmpDevice, i);
+		CUcontext tmpContext;
+		cuCtxCreate(&tmpContext, 0, tmpDevice);
+		if(cuMemGetInfo(&free, &total)) exit(1);
+		cuCtxDestroy(tmpContext);
+		if (DebugOutput >= 1) printf("Obtained current memory usage for device %d\n", i);
+		checkCudaErrors(cudaGetDeviceProperties(&deviceProp, i));
+		if (DebugOutput >= 1) printf("Obtained device properties for device %d\n", i);
+
+		if (DebugOutput >= 1) printf("%2d: %s (Rev: %d.%d - Mem Avail %lld / %lld)", i, deviceProp.name, deviceProp.major, deviceProp.minor, (long long int) free, (long long int) deviceProp.totalGlobalMem);
+		long long int deviceSpeed = (long long int) deviceProp.multiProcessorCount * (long long int) deviceProp.clockRate * (long long int) deviceProp.warpSize;
+		if (deviceSpeed > bestDeviceSpeed)
+		{
+			bestDevice = i;
+			bestDeviceSpeed = deviceSpeed;
+		}
+	}
+
+	cudaGetDeviceProperties(&deviceProp ,bestDevice); 
+
+	if (DebugOutput >= 1)
+	{
+		printf("Using CUDA Device %s with Properties:", deviceProp.name);
+		printf("totalGlobalMem = %lld", (unsigned long long int) deviceProp.totalGlobalMem);
+		printf("sharedMemPerBlock = %lld", (unsigned long long int) deviceProp.sharedMemPerBlock);
+		printf("regsPerBlock = %d", deviceProp.regsPerBlock);
+		printf("warpSize = %d", deviceProp.warpSize);
+		printf("memPitch = %lld", (unsigned long long int) deviceProp.memPitch);
+		printf("maxThreadsPerBlock = %d", deviceProp.maxThreadsPerBlock);
+		printf("maxThreadsDim = %d %d %d", deviceProp.maxThreadsDim[0], deviceProp.maxThreadsDim[1], deviceProp.maxThreadsDim[2]);
+		printf("maxGridSize = %d %d %d", deviceProp.maxGridSize[0], deviceProp.maxGridSize[1], deviceProp.maxGridSize[2]);
+		printf("totalConstMem = %lld", (unsigned long long int) deviceProp.totalConstMem);
+		printf("major = %d", deviceProp.major);
+		printf("minor = %d", deviceProp.minor);
+		printf("clockRate = %d", deviceProp.clockRate);
+		printf("memoryClockRate = %d", deviceProp.memoryClockRate);
+		printf("multiProcessorCount = %d", deviceProp.multiProcessorCount);
+		printf("textureAlignment = %lld", (unsigned long long int) deviceProp.textureAlignment);
+	}
+	
+	return(0);
+}
+
 int bioem_cuda::deviceInit()
 {
 	deviceExit();
+	
+
 
 	if (FFTAlgo) GPUAlgo = 2;
 
@@ -471,5 +540,14 @@ void bioem_cuda::free_device_host(void* ptr)
 
 bioem* bioem_cuda_create()
 {
+	int count;
+	
+	if (cudaGetDeviceCount(&count) != cudaSuccess) count = 0;
+	if (count == 0)
+	{
+		printf("No CUDA device available, using fallback to CPU version\n");
+		return new bioem;
+	}
+
 	return new bioem_cuda;
 }
diff --git a/include/bioem_cuda_internal.h b/include/bioem_cuda_internal.h
index 67d1206fd8e4dc184e3183401f486f3f88f76d21..9aaa689d7aac7d445008cae1f2aaf2e6f5508ea1 100644
--- a/include/bioem_cuda_internal.h
+++ b/include/bioem_cuda_internal.h
@@ -26,6 +26,9 @@ protected:
 	virtual int deviceStartRun();
 	virtual int deviceFinishRun();
 	int deviceExit();
+	
+private:
+	int selectCudaDevice();
 
 	int deviceInitialized;