diff --git a/CMakeLists.txt b/CMakeLists.txt
index 52610ac89..76d84fa0a 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -478,6 +478,8 @@ if (WITH_CUDA)
     # Keep deprecated but not yet dropped Compute Capabilities.
     if(CUDA_VERSION_MAJOR EQUAL 11)
       list(INSERT CUDA_ARCH_LIST 0 "3.5" "5.0")
+    elseif(CUDA_VERSION_MAJOR EQUAL 12)
+      list(INSERT CUDA_ARCH_LIST 0 "5.0")
     endif()
     list(REMOVE_DUPLICATES CUDA_ARCH_LIST)
   endif()
diff --git a/src/ops/awq/dequantize_gpu.cu b/src/ops/awq/dequantize_gpu.cu
index f63361035..baf62af36 100644
--- a/src/ops/awq/dequantize_gpu.cu
+++ b/src/ops/awq/dequantize_gpu.cu
@@ -13,6 +13,9 @@ namespace ctranslate2 {
                                                               int in_c,
                                                               int out_c)
      {
+#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ < 530
+       assert(false);
+#else
        if (blockIdx.z > 0) {
          B = B + blockIdx.z * in_c * out_c / 8;
          scaling_factors = scaling_factors + blockIdx.z * in_c * out_c / G;
@@ -60,6 +63,7 @@ namespace ctranslate2 {
        for (int i=0; i<8; ++i) {
          *(C_ptr2 + i) = B_shared[i];
        }
+#endif
      }
 
      template <Device D, typename InT, typename OutT>