diff --git a/mlx/backend/cuda/CMakeLists.txt b/mlx/backend/cuda/CMakeLists.txt
index d5041b2ae..e4f36074a 100644
--- a/mlx/backend/cuda/CMakeLists.txt
+++ b/mlx/backend/cuda/CMakeLists.txt
@@ -44,12 +44,16 @@ target_compile_options(mlx
 # CUDA 12.8 emits warning #20280-D for copy kernels which is a false positive.
 # Explicitly pass this flag to suppress the warning, it is safe to set it to
 # true but the warning wouldn't be suppressed.
-if(CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL 12.8)
+if(CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL 12.8.0)
   target_compile_options(
     mlx
     PRIVATE "$<$<COMPILE_LANGUAGE:CUDA>:--static-global-template-stub=false>")
 endif()
 
+# Suppress warning when building for compute capability 7 used by V100.
+target_compile_options(
+  mlx PRIVATE "$<$<COMPILE_LANGUAGE:CUDA>:--Wno-deprecated-gpu-targets>")
+
 # Compute capability 7 is required for synchronization between CPU/GPU with
 # managed memory. TODO: Add more architectures for potential performance gain.
 set(MLX_CUDA_ARCHITECTURES