diff --git a/intern/cycles/blender/blender_sync.cpp b/intern/cycles/blender/blender_sync.cpp
index b66004bee58..4c8fe350b88 100644
--- a/intern/cycles/blender/blender_sync.cpp
+++ b/intern/cycles/blender/blender_sync.cpp
@@ -257,7 +257,7 @@ SessionParams BlenderSync::get_session_params(BL::Scene b_scene, bool background
 	PointerRNA cscene = RNA_pointer_get(&b_scene.ptr, "cycles");
 
 	/* feature set */
-	bool experimental = (RNA_enum_get(&cscene, "feature_set") != 0);
+	params.experimental = (RNA_enum_get(&cscene, "feature_set") != 0);
 
 	/* device type */
 	params.device_type = DEVICE_CPU;
@@ -266,14 +266,14 @@ SessionParams BlenderSync::get_session_params(BL::Scene b_scene, bool background
 		vector<DeviceType> types = Device::available_types();
 		DeviceType dtype;
 		
-		if(!experimental || RNA_enum_get(&cscene, "gpu_type") == 0)
+		if(!params.experimental || RNA_enum_get(&cscene, "gpu_type") == 0)
 			dtype = DEVICE_CUDA;
 		else
 			dtype = DEVICE_OPENCL;
 
 		if(device_type_available(types, dtype))
 			params.device_type = dtype;
-		else if(experimental && device_type_available(types, DEVICE_OPENCL))
+		else if(params.experimental && device_type_available(types, DEVICE_OPENCL))
 			params.device_type = DEVICE_OPENCL;
 		else if(device_type_available(types, DEVICE_CUDA))
 			params.device_type = DEVICE_CUDA;
diff --git a/intern/cycles/device/device.h b/intern/cycles/device/device.h
index 5552e6ab7e2..5b87b11b6b8 100644
--- a/intern/cycles/device/device.h
+++ b/intern/cycles/device/device.h
@@ -112,7 +112,7 @@ public:
 	virtual void *osl_memory() { return NULL; }
 
 	/* load/compile kernels, must be called before adding tasks */ 
-	virtual bool load_kernels() { return true; }
+	virtual bool load_kernels(bool experimental) { return true; }
 
 	/* tasks */
 	virtual void task_add(DeviceTask& task) = 0;
diff --git a/intern/cycles/device/device_cuda.cpp b/intern/cycles/device/device_cuda.cpp
index 83d3f79d96f..eb0e20db355 100644
--- a/intern/cycles/device/device_cuda.cpp
+++ b/intern/cycles/device/device_cuda.cpp
@@ -214,6 +214,21 @@ public:
 		return string("CUDA ") + deviceName;
 	}
 
+	bool support_device(bool experimental)
+	{
+		if(!experimental) {
+			int major, minor;
+			cuDeviceComputeCapability(&major, &minor, cuDevId);
+
+			if(major <= 1 && minor <= 2) {
+				cuda_error(string_printf("CUDA device supported only with shader model 1.3 or up, found %d.%d.", major, minor));
+				return false;
+			}
+		}
+
+		return true;
+	}
+
 	string compile_kernel()
 	{
 		/* compute cubin name */
@@ -236,11 +251,11 @@ public:
 		if(path_exists(cubin))
 			return cubin;
 
-#ifdef WITH_CUDA_BINARIES
+#if defined(WITH_CUDA_BINARIES) && defined(_WIN32)
 		if(major <= 1 && minor <= 2)
 			cuda_error(string_printf("CUDA device supported only with shader model 1.3 or up, found %d.%d.", major, minor));
 		else
-			cuda_error("CUDA binary kernel for this graphics card not found.");
+			cuda_error("CUDA binary kernel for this graphics card shader model (%d.%d) not found.", major, minor);
 		return "";
 #else
 		/* if not, find CUDA compiler */
@@ -283,12 +298,15 @@ public:
 #endif
 	}
 
-	bool load_kernels()
+	bool load_kernels(bool experimental)
 	{
 		/* check if cuda init succeeded */
 		if(cuContext == 0)
 			return false;
 
+		if(!support_device(experimental))
+			return false;
+
 		/* get kernel */
 		string cubin = compile_kernel();
 
diff --git a/intern/cycles/device/device_multi.cpp b/intern/cycles/device/device_multi.cpp
index 128c80ac396..fc5348ad168 100644
--- a/intern/cycles/device/device_multi.cpp
+++ b/intern/cycles/device/device_multi.cpp
@@ -132,10 +132,10 @@ public:
 		return desc.str();
 	}
 
-	bool load_kernels()
+	bool load_kernels(bool experimental)
 	{
 		foreach(SubDevice& sub, devices)
-			if(!sub.device->load_kernels())
+			if(!sub.device->load_kernels(experimental))
 				return false;
 
 		return true;
diff --git a/intern/cycles/device/device_opencl.cpp b/intern/cycles/device/device_opencl.cpp
index d783ae6c174..8eaaebc6629 100644
--- a/intern/cycles/device/device_opencl.cpp
+++ b/intern/cycles/device/device_opencl.cpp
@@ -365,7 +365,7 @@ public:
 		return md5.get_hex();
 	}
 
-	bool load_kernels()
+	bool load_kernels(bool experimental)
 	{
 		/* verify if device was initialized */
 		if(!device_initialized) {
diff --git a/intern/cycles/render/session.cpp b/intern/cycles/render/session.cpp
index 73ac033623d..42b4a2bb7e4 100644
--- a/intern/cycles/render/session.cpp
+++ b/intern/cycles/render/session.cpp
@@ -410,7 +410,7 @@ void Session::run()
 	/* load kernels */
 	progress.set_status("Loading render kernels (may take a few minutes the first time)");
 
-	if(!device->load_kernels()) {
+	if(!device->load_kernels(params.experimental)) {
 		string message = device->error_message();
 		if(message == "")
 			message = "Failed loading render kernel, see console for errors";
diff --git a/intern/cycles/render/session.h b/intern/cycles/render/session.h
index 82c14227d03..ce7f420096a 100644
--- a/intern/cycles/render/session.h
+++ b/intern/cycles/render/session.h
@@ -43,6 +43,7 @@ public:
 	string output_path;
 
 	bool progressive;
+	bool experimental;
 	int samples;
 	int tile_size;
 	int min_size;
@@ -59,6 +60,7 @@ public:
 		output_path = "";
 
 		progressive = false;
+		experimental = false;
 		samples = INT_MAX;
 		tile_size = 64;
 		min_size = 64;
@@ -75,6 +77,7 @@ public:
 		&& output_path == params.output_path
 		/* && samples == params.samples */
 		&& progressive == params.progressive
+		&& experimental == params.experimental
 		&& tile_size == params.tile_size
 		&& min_size == params.min_size
 		&& threads == params.threads