diff --git a/conda/pytorch-nightly/build.sh b/conda/pytorch-nightly/build.sh
index eee3a3d86..fe76cb828 100755
--- a/conda/pytorch-nightly/build.sh
+++ b/conda/pytorch-nightly/build.sh
@@ -57,12 +57,12 @@ if [[ -n "$build_with_cuda" ]]; then
     export USE_CUSPARSELT=1 # link with cusparselt
 
     if [[ $CUDA_VERSION == 11.8* ]]; then
-        TORCH_CUDA_ARCH_LIST="$TORCH_CUDA_ARCH_LIST;3.7+PTX;9.0"
+        TORCH_CUDA_ARCH_LIST="$TORCH_CUDA_ARCH_LIST;3.7+PTX;9.0a"
         #for cuda 11.8 include all dynamic loading libraries
         DEPS_LIST=(/usr/local/cuda/lib64/libcudnn*.so.9 /usr/local/cuda-11.8/extras/CUPTI/lib64/libcupti.so.11.8 /usr/local/cuda/lib64/libcusparseLt.so.0)
     elif [[ $CUDA_VERSION == 12.1* || $CUDA_VERSION == 12.4* ]]; then
         # cuda 12 does not support sm_3x
-        TORCH_CUDA_ARCH_LIST="$TORCH_CUDA_ARCH_LIST;9.0"
+        TORCH_CUDA_ARCH_LIST="$TORCH_CUDA_ARCH_LIST;9.0a"
         # for cuda 12.1 (12.4) we use cudnn 9.1 and include all dynamic loading libraries
         DEPS_LIST=(/usr/local/cuda/lib64/libcudnn*.so.9 /usr/local/cuda/extras/CUPTI/lib64/libcupti.so.12 /usr/local/cuda/lib64/libcusparseLt.so.0)
     fi
diff --git a/manywheel/build_cuda.sh b/manywheel/build_cuda.sh
index 1d8e8b295..1aaf53680 100644
--- a/manywheel/build_cuda.sh
+++ b/manywheel/build_cuda.sh
@@ -61,18 +61,18 @@ TORCH_CUDA_ARCH_LIST="5.0;6.0;7.0;7.5;8.0;8.6"
 case ${CUDA_VERSION} in
     12.4)
         if [[ "$GPU_ARCH_TYPE" = "cuda-aarch64" ]]; then
-            TORCH_CUDA_ARCH_LIST="9.0"
+            TORCH_CUDA_ARCH_LIST="9.0a"
         else
-            TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};9.0"
+            TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};9.0a"
         fi
         EXTRA_CAFFE2_CMAKE_FLAGS+=("-DATEN_NO_TEST=ON")
         ;;
     12.1)
-        TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};9.0"
+        TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};9.0a"
         EXTRA_CAFFE2_CMAKE_FLAGS+=("-DATEN_NO_TEST=ON")
         ;;
     11.8)
-        TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};3.7;9.0"
+        TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};3.7;9.0a"
         EXTRA_CAFFE2_CMAKE_FLAGS+=("-DATEN_NO_TEST=ON")
         ;;
     11.[67])
diff --git a/windows/cuda121.bat b/windows/cuda121.bat
index e8be06baa..86775943a 100644
--- a/windows/cuda121.bat
+++ b/windows/cuda121.bat
@@ -37,10 +37,10 @@ IF "%CUDA_PATH_V121%"=="" (
 )
 
 IF "%BUILD_VISION%" == "" (
-    set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0
+    set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0a
     set TORCH_NVCC_FLAGS=-Xfatbin -compress-all
 ) ELSE (
-    set NVCC_FLAGS=-D__CUDA_NO_HALF_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_50,code=sm_50 -gencode=arch=compute_60,code=sm_60 -gencode=arch=compute_70,code=sm_70 -gencode=arch=compute_75,code=sm_75 -gencode=arch=compute_80,code=compute_80 -gencode=arch=compute_86,code=compute_86 -gencode=arch=compute_90,code=compute_90
+    set NVCC_FLAGS=-D__CUDA_NO_HALF_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_50,code=sm_50 -gencode=arch=compute_60,code=sm_60 -gencode=arch=compute_70,code=sm_70 -gencode=arch=compute_75,code=sm_75 -gencode=arch=compute_80,code=compute_80 -gencode=arch=compute_86,code=compute_86 -gencode=arch=compute_90a,code=compute_90a
 )
 
 set "CUDA_PATH=%CUDA_PATH_V121%"
diff --git a/windows/cuda124.bat b/windows/cuda124.bat
index b9594e159..5530eaef7 100644
--- a/windows/cuda124.bat
+++ b/windows/cuda124.bat
@@ -37,10 +37,10 @@ IF "%CUDA_PATH_V124%"=="" (
 )
 
 IF "%BUILD_VISION%" == "" (
-    set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0
+    set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0a
     set TORCH_NVCC_FLAGS=-Xfatbin -compress-all
 ) ELSE (
-    set NVCC_FLAGS=-D__CUDA_NO_HALF_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_50,code=sm_50 -gencode=arch=compute_60,code=sm_60 -gencode=arch=compute_70,code=sm_70 -gencode=arch=compute_75,code=sm_75 -gencode=arch=compute_80,code=compute_80 -gencode=arch=compute_86,code=compute_86 -gencode=arch=compute_90,code=compute_90
+    set NVCC_FLAGS=-D__CUDA_NO_HALF_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_50,code=sm_50 -gencode=arch=compute_60,code=sm_60 -gencode=arch=compute_70,code=sm_70 -gencode=arch=compute_75,code=sm_75 -gencode=arch=compute_80,code=compute_80 -gencode=arch=compute_86,code=compute_86 -gencode=arch=compute_90a,code=compute_90a
 )
 
 set "CUDA_PATH=%CUDA_PATH_V124%"