diff --git a/intern/cycles/app/CMakeLists.txt b/intern/cycles/app/CMakeLists.txt
index a8e0297aeaa..d1f86a5fe7d 100644
--- a/intern/cycles/app/CMakeLists.txt
+++ b/intern/cycles/app/CMakeLists.txt
@@ -126,7 +126,8 @@ if(WITH_CYCLES_CUBIN_COMPILER)
 	# though we are building 32 bit blender a 64 bit cubin_cc will have
 	# to be build to compile the cubins.
 	if(MSVC AND NOT CMAKE_CL_64)
-		Message("cycles_cubin_cc not supported on x86")
+		message("Building with CUDA not supported on 32 bit, skipped")
+		set(WITH_CYCLES_CUDA_BINARIES OFF)
 	else()
 		set(SRC
 			cycles_cubin_cc.cpp
diff --git a/intern/cycles/device/device_cuda.cpp b/intern/cycles/device/device_cuda.cpp
index d28080c667a..5703aa7144e 100644
--- a/intern/cycles/device/device_cuda.cpp
+++ b/intern/cycles/device/device_cuda.cpp
@@ -2479,7 +2479,7 @@ void device_cuda_info(vector<DeviceInfo>& devices)
 		cuDeviceGetAttribute(&major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, num);
 		if(major < 3) {
 			VLOG(1) << "Ignoring device \"" << name
-			        << "\", compute capability is too low.";
+			        << "\", this graphics card is no longer supported.";
 			continue;
 		}
 
diff --git a/intern/cycles/kernel/CMakeLists.txt b/intern/cycles/kernel/CMakeLists.txt
index 4d4b8602c78..50ea03a1f8f 100644
--- a/intern/cycles/kernel/CMakeLists.txt
+++ b/intern/cycles/kernel/CMakeLists.txt
@@ -416,9 +416,13 @@ if(WITH_CYCLES_CUDA_BINARIES)
 	endmacro()
 
 	foreach(arch ${CYCLES_CUDA_BINARIES_ARCH})
-		# Compile regular kernel
-		CYCLES_CUDA_KERNEL_ADD(${arch} filter "" "${cuda_filter_sources}" FALSE)
-		CYCLES_CUDA_KERNEL_ADD(${arch} kernel "" "${cuda_sources}" FALSE)
+		if(${arch} MATCHES "sm_2.")
+			message(STATUS "CUDA binaries for ${arch} are no longer supported, skipped.")
+		else()
+			# Compile regular kernel
+			CYCLES_CUDA_KERNEL_ADD(${arch} filter "" "${cuda_filter_sources}" FALSE)
+			CYCLES_CUDA_KERNEL_ADD(${arch} kernel "" "${cuda_sources}" FALSE)
+		endif()
 
 		if(WITH_CYCLES_CUDA_SPLIT_KERNEL_BINARIES)
 			# Compile split kernel