diff -ruN
--- lib/CUDAKernels/kernels.cu
+++
@@ -206,7 +206,7 @@
 // template<>           __device__ __forceinline__  double RSQRT(double val) { return 1.0/sqrt(val); }
 
 
-
+#if (CUDART_VERSION <= 8000)
 __device__ double atomicAdd(double* address, double val)
 {
     unsigned long long int* address_as_ull =
@@ -220,7 +220,7 @@
     } while (assumed != old);
     return __longlong_as_double(old);
 }
-
+#endif
 
 __device__ __forceinline__ double atomicMin(double *address, double val)
 {
diff -ruN
--- lib/Makefile
+++
@@ -5,7 +5,8 @@
 
 .SUFFIXES: .o .cpp .ptx .cu
 
-CUDA_TK  ?= /usr/local/cuda
+CUDAHOME ?= /usr/local/cuda
+CUDA_TK   = $(CUDAHOME)
 
 
 testRunFlags1=
@@ -36,17 +37,8 @@
 NVCC      = $(CUDA_TK)/bin/nvcc
 
 
-# Support older CUDA versions out of the box
-NVCCVERSION=$(shell "${NVCC}" --version | grep ^Cuda | sed 's/^.* //g')
-ifeq "${NVCCVERSION}" "V5.5.22"
-	NVCCFLAGS ?= -arch sm_20
-else
-	NVCCFLAGS ?= -arch sm_30
-endif
-
-#NVCCFLAGS = -arch sm_35
-#NVCCFLAGS ?= -arch sm_30
-#NVCCFLAGS = -arch sm_20
+GPUARCH   ?= sm_75
+NVCCFLAGS  = -arch $(GPUARCH)
 NVCCFLAGS +=  ${testRunFlags}
 
 # Use with Mac OS X