diff --git a/cpu-miner.c b/cpu-miner.c
index 01100be..ef3efa4 100755
--- a/cpu-miner.c
+++ b/cpu-miner.c
@@ -53,7 +53,8 @@ extern "C"
 {
 #endif
 	int cuda_num_devices();
-	void cuda_deviceinfo();
+	void cuda_deviceinfo(int);
+	void cuda_set_device_config(int);
 	int cuda_finddevice(char *name);
 #ifdef __cplusplus
 }
@@ -1452,34 +1453,28 @@ void parse_device_config(int device, char *config, int *blocks, int *threads)
 	char *p;
 	int tmp_blocks, tmp_threads;
 
-	if(config == NULL) goto usedefault;
-
+	if(config == NULL)
+		return;
 
 	p = strtok(config, "x");
 	if(!p)
-		goto usedefault;
+		return;
 
 	tmp_threads = atoi(p);
 	if(tmp_threads < 4 || tmp_threads > 1024)
-		goto usedefault;
+		return;
 
 	p = strtok(NULL, "x");
 	if(!p)
-		goto usedefault;
+		return;
 
 	tmp_blocks = atoi(p);
 	if(tmp_blocks < 1)
-		goto usedefault;
+		return;
 
 	*blocks = tmp_blocks;
 	*threads = tmp_threads;
 	return;
-
-usedefault:
-	*blocks = 4 * device_mpcount[device];
-	*threads = opt_cn_threads;
-	return;
-
 }
 
 static void parse_arg(int key, char *arg)
@@ -1943,7 +1938,9 @@ int main(int argc, char *argv[])
 	/* parse command line */
 	parse_cmdline(argc, argv);
 	color_init();
-	cuda_deviceinfo();
+
+	cuda_deviceinfo(num_processors);
+	cuda_set_device_config(num_processors);
 
 	if(!opt_benchmark && !rpc_url)
 	{
diff --git a/cryptonight/cryptonight.cu b/cryptonight/cryptonight.cu
index 1cb9ba4..485815c 100755
--- a/cryptonight/cryptonight.cu
+++ b/cryptonight/cryptonight.cu
@@ -47,69 +47,73 @@ extern "C" int cuda_num_devices()
 	}
 	return GPU_N;
 }
-
-extern "C" void cuda_deviceinfo()
+extern "C" void cuda_set_device_config(int GPU_N)
 {
-	cudaError_t err;
-	int GPU_N;
-	err = cudaGetDeviceCount(&GPU_N);
-	if(err != cudaSuccess)
-	{
-		if(err != cudaErrorNoDevice)
-			applog(LOG_ERR, "No CUDA device found!");
-		else
-			applog(LOG_ERR, "Unable to query number of CUDA devices!");
-		exit(1);
-	}
-
 	for(int i = 0; i < GPU_N; i++)
 	{
-		cudaDeviceProp props;
-		cudaError_t err = cudaGetDeviceProperties(&props, device_map[i]);
-		if(err != cudaSuccess)
+		if(device_config[i][0] == 0)
 		{
-			printf("\nGPU %d: %s\n%s line %d\n", device_map[i], cudaGetErrorString(err), __FILE__, __LINE__);
-			exit(1);
-		}
-
-		device_name[i] = strdup(props.name);
-		device_mpcount[i] = props.multiProcessorCount;
-		device_arch[i][0] = props.major;
-		device_arch[i][1] = props.minor;
-
-		device_config[i][0] = props.multiProcessorCount * (props.major < 3 ? 2 : 3);
-		device_config[i][1] = 64;
+			device_config[i][0] = device_mpcount[i] * 4;
+			device_config[i][1] = 64;
 
-		/* sm_20 devices can only run 512 threads per cuda block
-		* `cryptonight_core_gpu_phase1` and `cryptonight_core_gpu_phase3` starts
-		* `8 * ctx->device_threads` threads per block
-		*/
-		if(props.major < 6) {
-
-			//Try to stay under 950 threads ( 1900MiB memory per for hashes )
-			while(device_config[i][0] * device_config[i][1] >= 950 && device_config[i][1] > 2)
+			/* sm_20 devices can only run 512 threads per cuda block
+			* `cryptonight_core_gpu_phase1` and `cryptonight_core_gpu_phase3` starts
+			* `8 * ctx->device_threads` threads per block
+			*/
+			if(device_arch[i][0] < 6)
 			{
-				device_config[i][1] /= 2;
+				//Try to stay under 950 threads ( 1900MiB memory per for hashes )
+				while(device_config[i][0] * device_config[i][1] >= 950 && device_config[i][1] > 2)
+				{
+					device_config[i][1] /= 2;
+				}
 			}
-
 			//Stay within 85% of the available RAM
 			while(device_config[i][1] > 2)
 			{
 				size_t freeMemory = 0;
 				size_t totalMemoery = 0;
 
-				cudaMemGetInfo(&freeMemory, &totalMemoery);
-				freeMemory = (freeMemory * size_t(85)) / 100;
+				cudaError_t err = cudaMemGetInfo(&freeMemory, &totalMemoery);
+				if(err == cudaSuccess)
+				{
+					freeMemory = (freeMemory * size_t(85)) / 100;
 
-				if(freeMemory > size_t(device_config[i][0]) * size_t(device_config[i][1]) * size_t(2u * 1024u * 1024u)) {
-					break;
-				} else {
-					device_config[i][1] /= 2;
+					if(freeMemory > size_t(device_config[i][0]) * size_t(device_config[i][1]) * 2097832)
+					{
+						break;
+					}
+					else
+					{
+						device_config[i][1] /= 2;
+					}
 				}
+				else
+					applog(LOG_WARNING, "GPU #%d: CUDA error: %s", device_map[i], cudaGetErrorString(err));
 			}
 		}
 	}
 }
+extern "C" int cuda_deviceinfo(int GPU_N)
+{
+	cudaError_t err;
+	for(int i = 0; i < GPU_N; i++)
+	{
+		cudaDeviceProp props;
+		cudaError_t err = cudaGetDeviceProperties(&props, device_map[i]);
+		if(err != cudaSuccess)
+		{
+			printf("\nGPU %d: %s\n%s line %d\n", device_map[i], cudaGetErrorString(err), __FILE__, __LINE__);
+			exit(1);
+		}
+
+		device_name[i] = strdup(props.name);
+		device_mpcount[i] = props.multiProcessorCount;
+		device_arch[i][0] = props.major;
+		device_arch[i][1] = props.minor;
+	}
+	return GPU_N;
+}
 
 static bool substringsearch(const char *haystack, const char *needle, int &match)
 {