diff --git a/opm/simulators/linalg/bda/cusparseSolverBackend.cu b/opm/simulators/linalg/bda/cusparseSolverBackend.cu
index 65ef8e2cb..3a29cd573 100644
--- a/opm/simulators/linalg/bda/cusparseSolverBackend.cu
+++ b/opm/simulators/linalg/bda/cusparseSolverBackend.cu
@@ -227,8 +227,8 @@ void cusparseSolverBackend<block_size>::initialize(int N, int nnz, int dim) {
     cudaMalloc((void**)&d_t, sizeof(double) * N);
     cudaMalloc((void**)&d_v, sizeof(double) * N);
     cudaMalloc((void**)&d_bVals, sizeof(double) * nnz);
-    cudaMalloc((void**)&d_bCols, sizeof(double) * nnz);
-    cudaMalloc((void**)&d_bRows, sizeof(double) * (Nb + 1));
+    cudaMalloc((void**)&d_bCols, sizeof(int) * nnzb);
+    cudaMalloc((void**)&d_bRows, sizeof(int) * (Nb + 1));
     cudaMalloc((void**)&d_mVals, sizeof(double) * nnz);
     cudaCheckLastError("Could not allocate enough memory on GPU");
 
@@ -295,7 +295,7 @@ void cusparseSolverBackend<block_size>::copy_system_to_gpu(double *vals, int *ro
     cudaMemcpyAsync(d_bVals, vals, nnz * sizeof(double), cudaMemcpyHostToDevice, stream);
 #endif
 
-    cudaMemcpyAsync(d_bCols, cols, nnz * sizeof(int), cudaMemcpyHostToDevice, stream);
+    cudaMemcpyAsync(d_bCols, cols, nnzb * sizeof(int), cudaMemcpyHostToDevice, stream);
     cudaMemcpyAsync(d_bRows, rows, (Nb + 1) * sizeof(int), cudaMemcpyHostToDevice, stream);
     cudaMemcpyAsync(d_b, b, N * sizeof(double), cudaMemcpyHostToDevice, stream);
     cudaMemsetAsync(d_x, 0, sizeof(double) * N, stream);
diff --git a/opm/simulators/linalg/bda/cusparseSolverBackend.hpp b/opm/simulators/linalg/bda/cusparseSolverBackend.hpp
index 838446826..cec9618af 100644
--- a/opm/simulators/linalg/bda/cusparseSolverBackend.hpp
+++ b/opm/simulators/linalg/bda/cusparseSolverBackend.hpp
@@ -84,7 +84,7 @@ private:
     /// Copy linear system to GPU
     /// \param[in] vals        array of nonzeroes, each block is stored row-wise, contains nnz values
     /// \param[in] rows        array of rowPointers, contains N/dim+1 values
-    /// \param[in] cols        array of columnIndices, contains nnz values
+    /// \param[in] cols        array of columnIndices, contains nnzb values
     /// \param[in] b           input vector, contains N values
     void copy_system_to_gpu(double *vals, int *rows, int *cols, double *b);