diff --git a/.github/scripts/build.sh b/.github/scripts/build.sh
index 4e5ae4aa9a..419b0a465e 100755
--- a/.github/scripts/build.sh
+++ b/.github/scripts/build.sh
@@ -50,7 +50,6 @@ git submodule sync && git submodule update --init --recursive
 python -m pip install -r requirements.txt
 python -m pip install mkl-static mkl-include
 export USE_STATIC_MKL=1
-export USE_ONEMKL=1
 export USE_XCCL=1
 export PYTORCH_EXTRA_INSTALL_REQUIREMENTS=" \
     intel-cmplr-lib-rt==2025.1.1 | \
diff --git a/.github/workflows/_windows_ut.yml b/.github/workflows/_windows_ut.yml
index 6c168870a8..3c9d542a8f 100644
--- a/.github/workflows/_windows_ut.yml
+++ b/.github/workflows/_windows_ut.yml
@@ -115,7 +115,6 @@ jobs:
           pip install cmake setuptools==72.1.0 clang-format
           pip install mkl-static mkl-include
           set USE_STATIC_MKL=1
-          set USE_ONEMKL=1
           copy "%CONDA_PREFIX%\Library\bin\libiomp*5md.dll" .\torch\lib
           copy "%CONDA_PREFIX%\Library\bin\uv.dll" .\torch\lib
           if defined CMAKE_PREFIX_PATH (
diff --git a/cmake/ONEMKL.cmake b/cmake/ONEMKL.cmake
index 04518f7a6f..aa04d4b0d7 100644
--- a/cmake/ONEMKL.cmake
+++ b/cmake/ONEMKL.cmake
@@ -1,12 +1,12 @@
-option(USE_ONEMKL "Build with ONEMKL XPU support" OFF)
+option(USE_ONEMKL_XPU "Build with ONEMKL XPU support" ON)
 
-if(DEFINED ENV{USE_ONEMKL})
-  set(USE_ONEMKL $ENV{USE_ONEMKL})
+if(DEFINED ENV{USE_ONEMKL_XPU})
+  set(USE_ONEMKL_XPU $ENV{USE_ONEMKL_XPU})
 endif()
 
-message(STATUS "USE_ONEMKL is set to ${USE_ONEMKL}")
+message(STATUS "USE_ONEMKL_XPU is set to ${USE_ONEMKL_XPU}")
 
-if(NOT USE_ONEMKL)
+if(NOT USE_ONEMKL_XPU)
   return()
 endif()
 
diff --git a/src/ATen/native/xpu/SpectralOps.cpp b/src/ATen/native/xpu/SpectralOps.cpp
index 7ed5464288..16f5a92987 100644
--- a/src/ATen/native/xpu/SpectralOps.cpp
+++ b/src/ATen/native/xpu/SpectralOps.cpp
@@ -1,11 +1,11 @@
-#if defined(USE_ONEMKL)
+#if defined(USE_ONEMKL_XPU)
 #include <ATen/native/xpu/mkl/SpectralOps.h>
 #else
 #include <ATen/native/Resize.h>
 #include <ATen/ops/_fft_c2c_native.h>
 #include <ATen/ops/_fft_c2r_native.h>
 #include <ATen/ops/_fft_r2c_native.h>
-#endif // USE_ONEMKL
+#endif // USE_ONEMKL_XPU
 
 namespace at::native {
 
@@ -16,13 +16,13 @@ Tensor _fft_c2c_xpu(
     bool forward) {
   TORCH_CHECK(self.is_complex());
 
-#if defined(USE_ONEMKL)
+#if defined(USE_ONEMKL_XPU)
   return native::xpu::_fft_c2c_mkl(self, dim, normalization, forward);
 #else
   Tensor out_cpu = native::_fft_c2c_mkl(
       self.to(Device(at::kCPU)), dim, normalization, forward);
   return out_cpu.to(Device(at::kXPU));
-#endif // USE_ONEMKL
+#endif // USE_ONEMKL_XPU
 }
 
 Tensor& _fft_c2c_xpu_out(
@@ -33,7 +33,7 @@ Tensor& _fft_c2c_xpu_out(
     Tensor& out) {
   TORCH_CHECK(self.is_complex());
 
-#if defined(USE_ONEMKL)
+#if defined(USE_ONEMKL_XPU)
   return native::xpu::_fft_c2c_mkl_out(self, dim, normalization, forward, out);
 #else
   Tensor out_cpu = native::_fft_c2c_mkl(
@@ -41,7 +41,7 @@ Tensor& _fft_c2c_xpu_out(
   at::native::resize_output(out, out_cpu.sizes());
   out.copy_(out_cpu);
   return out;
-#endif // USE_ONEMKL
+#endif // USE_ONEMKL_XPU
 }
 
 Tensor _fft_c2r_xpu(
@@ -51,13 +51,13 @@ Tensor _fft_c2r_xpu(
     int64_t last_dim_size) {
   TORCH_CHECK(self.is_complex());
 
-#if defined(USE_ONEMKL)
+#if defined(USE_ONEMKL_XPU)
   return native::xpu::_fft_c2r_mkl(self, dim, normalization, last_dim_size);
 #else
   Tensor out_cpu = native::_fft_c2r_mkl(
       self.to(Device(at::kCPU)), dim, normalization, last_dim_size);
   return out_cpu.to(Device(at::kXPU));
-#endif // USE_ONEMKL
+#endif // USE_ONEMKL_XPU
 }
 
 Tensor& _fft_c2r_xpu_out(
@@ -68,7 +68,7 @@ Tensor& _fft_c2r_xpu_out(
     Tensor& out) {
   TORCH_CHECK(self.is_complex());
 
-#if defined(USE_ONEMKL)
+#if defined(USE_ONEMKL_XPU)
   return native::xpu::_fft_c2r_mkl_out(
       self, dim, normalization, last_dim_size, out);
 #else
@@ -77,7 +77,7 @@ Tensor& _fft_c2r_xpu_out(
   at::native::resize_output(out, out_cpu.sizes());
   out.copy_(out_cpu);
   return out;
-#endif // USE_ONEMKL
+#endif // USE_ONEMKL_XPU
 }
 
 Tensor _fft_r2c_xpu(
@@ -87,13 +87,13 @@ Tensor _fft_r2c_xpu(
     bool onesided) {
   TORCH_CHECK(self.is_floating_point());
 
-#if defined(USE_ONEMKL)
+#if defined(USE_ONEMKL_XPU)
   return native::xpu::_fft_r2c_mkl(self, dim, normalization, onesided);
 #else
   Tensor out_cpu = native::_fft_r2c_mkl(
       self.to(Device(at::kCPU)), dim, normalization, onesided);
   return out_cpu.to(Device(at::kXPU));
-#endif // USE_ONEMKL
+#endif // USE_ONEMKL_XPU
 }
 
 Tensor& _fft_r2c_xpu_out(
@@ -104,7 +104,7 @@ Tensor& _fft_r2c_xpu_out(
     Tensor& out) {
   TORCH_CHECK(self.is_floating_point());
 
-#if defined(USE_ONEMKL)
+#if defined(USE_ONEMKL_XPU)
   return native::xpu::_fft_r2c_mkl_out(self, dim, normalization, onesided, out);
 #else
   Tensor out_cpu = native::_fft_r2c_mkl(
@@ -112,7 +112,7 @@ Tensor& _fft_r2c_xpu_out(
   at::native::resize_output(out, out_cpu.sizes());
   out.copy_(out_cpu);
   return out;
-#endif // USE_ONEMKL
+#endif // USE_ONEMKL_XPU
 }
 
 } // namespace at::native
diff --git a/src/ATen/native/xpu/mkl/SpectralOps.cpp b/src/ATen/native/xpu/mkl/SpectralOps.cpp
index 4f1e028b4a..372cb2ed88 100644
--- a/src/ATen/native/xpu/mkl/SpectralOps.cpp
+++ b/src/ATen/native/xpu/mkl/SpectralOps.cpp
@@ -1,4 +1,4 @@
-#if defined(USE_ONEMKL)
+#if defined(USE_ONEMKL_XPU)
 #include <ATen/native/Resize.h>
 #include <ATen/native/SpectralOpsUtils.h>
 #include <ATen/native/xpu/mkl/SpectralOps.h>
@@ -591,4 +591,4 @@ Tensor& _fft_r2c_mkl_out(
 }
 
 } // namespace at::native::xpu
-#endif // USE_ONEMKL
+#endif // USE_ONEMKL_XPU
diff --git a/src/BuildOnLinux.cmake b/src/BuildOnLinux.cmake
index a946acae33..aee7118f01 100644
--- a/src/BuildOnLinux.cmake
+++ b/src/BuildOnLinux.cmake
@@ -145,8 +145,8 @@ foreach(lib ${TORCH_XPU_OPS_LIBRARIES})
   target_link_libraries(${lib} PUBLIC ${SYCL_LIBRARY})
 endforeach()
 
-if(USE_ONEMKL)
-  target_compile_options(torch_xpu_ops PRIVATE "-DUSE_ONEMKL")
+if(USE_ONEMKL_XPU)
+  target_compile_options(torch_xpu_ops PRIVATE "-DUSE_ONEMKL_XPU")
   target_include_directories(torch_xpu_ops PUBLIC ${TORCH_XPU_OPS_ONEMKL_INCLUDE_DIR})
   target_link_libraries(torch_xpu_ops PUBLIC ${TORCH_XPU_OPS_ONEMKL_LIBRARIES})
-endif()
\ No newline at end of file
+endif()
diff --git a/src/BuildOnWindows.cmake b/src/BuildOnWindows.cmake
index 665fe56f30..77bd353d61 100644
--- a/src/BuildOnWindows.cmake
+++ b/src/BuildOnWindows.cmake
@@ -342,8 +342,8 @@ foreach(lib ${TORCH_XPU_OPS_LIBRARIES})
   target_link_libraries(${lib} PUBLIC torch_cpu)
 endforeach()
 
-if(USE_ONEMKL)
-  target_compile_options(torch_xpu_ops PRIVATE "-DUSE_ONEMKL")
+if(USE_ONEMKL_XPU)
+  target_compile_options(torch_xpu_ops PRIVATE "-DUSE_ONEMKL_XPU")
   target_include_directories(torch_xpu_ops PUBLIC ${TORCH_XPU_OPS_ONEMKL_INCLUDE_DIR})
   target_link_libraries(torch_xpu_ops PUBLIC ${TORCH_XPU_OPS_ONEMKL_LIBRARIES})
-endif()
\ No newline at end of file
+endif()