diff --git a/tests/unit/test_tmem_cols.cu b/tests/unit/test_tmem_cols.cu
index a3dcaba5..e03190bc 100644
--- a/tests/unit/test_tmem_cols.cu
+++ b/tests/unit/test_tmem_cols.cu
@@ -74,7 +74,7 @@ int main() {
     float* d_out; cudaMalloc(&d_out, 2 * sizeof(float));
     cudaMemset(d_out, 0, 2 * sizeof(float));
     
-    test_tmem_2col<<<1, 32, 1024>>>(d_out);
+    test_tmem_2col<<<1, 64, 1024>>>(d_out);
     cudaError_t err = cudaDeviceSynchronize();
     if (err != cudaSuccess) { printf("CUDA ERROR: %s\n", cudaGetErrorString(err)); return 1; }