diff --git a/CMakeLists.txt b/CMakeLists.txt
index 4fcdf1eba09..f3a5655c5e1 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -171,7 +171,8 @@ cmake_dependent_option(
     USE_STATIC_CUDNN "Use cuDNN static libraries" OFF
     "USE_CUDNN" OFF)
 option(USE_FBGEMM "Use FBGEMM (quantized 8-bit server operators)" ON)
-option(USE_KINETO "Use Kineto profiling library" OFF)
+option(USE_KINETO "Use Kineto profiling library" ON)
+option(USE_CUPTI_SO "Use CUPTI as a shared library" OFF)
 option(USE_FAKELOWP "Use FakeLowp operators" OFF)
 option(USE_FFMPEG "Use ffmpeg" OFF)
 option(USE_GFLAGS "Use GFLAGS" OFF)
@@ -545,31 +546,12 @@ if(USE_FBGEMM AND ((CMAKE_SYSTEM_PROCESSOR STREQUAL "x86_64" AND CMAKE_SIZEOF_VO
   set(USE_FBGEMM OFF)
 endif()
 
-if(USE_KINETO AND INTERN_BUILD_MOBILE)
-  message(STATUS "Not using libkineto in a mobile build.")
-  set(USE_KINETO OFF)
-endif()
-
-if(USE_KINETO AND (NOT USE_CUDA))
-  message(STATUS "Not using libkineto in a non-CUDA build.")
-  set(USE_KINETO OFF)
-endif()
-
-if(USE_KINETO AND MSVC)
-  message(STATUS "Not using libkineto in a Windows build.")
-  set(USE_KINETO OFF)
-endif()
-
 include(cmake/Dependencies.cmake)
 
 if(USE_FBGEMM)
   string(APPEND CMAKE_CXX_FLAGS " -DUSE_FBGEMM")
 endif()
 
-if(USE_KINETO)
-  string(APPEND CMAKE_CXX_FLAGS " -DUSE_KINETO")
-endif()
-
 if(USE_QNNPACK)
   string(APPEND CMAKE_CXX_FLAGS " -DUSE_QNNPACK")
 endif()
diff --git a/cmake/Dependencies.cmake b/cmake/Dependencies.cmake
index e5fed395801..0ee503daa73 100644
--- a/cmake/Dependencies.cmake
+++ b/cmake/Dependencies.cmake
@@ -1787,43 +1787,68 @@ list(APPEND Caffe2_DEPENDENCY_LIBS fmt::fmt-header-only)
 set(BUILD_SHARED_LIBS ${TEMP_BUILD_SHARED_LIBS} CACHE BOOL "Build shared libs" FORCE)
 
 # ---[ Kineto
+if(USE_KINETO AND INTERN_BUILD_MOBILE)
+  message(STATUS "Not using libkineto in a mobile build.")
+  set(USE_KINETO OFF)
+endif()
+if(USE_KINETO AND (NOT USE_CUDA))
+  message(STATUS "Not using libkineto in a non-CUDA build.")
+  set(USE_KINETO OFF)
+endif()
+if(USE_KINETO AND MSVC)
+  message(STATUS "Not using libkineto in a Windows build.")
+  set(USE_KINETO OFF)
+endif()
 if(USE_KINETO)
-  if(USE_KINETO AND NOT TARGET kineto)
-    set(CAFFE2_THIRD_PARTY_ROOT "${PROJECT_SOURCE_DIR}/third_party" CACHE STRING "")
-    set(KINETO_SOURCE_DIR "${CAFFE2_THIRD_PARTY_ROOT}/kineto/libkineto" CACHE STRING "")
-    set(KINETO_BUILD_TESTS OFF CACHE BOOL "")
-    set(KINETO_LIBRARY_TYPE "static" CACHE STRING "")
-    set(CUDA_SOURCE_DIR "${CUDA_TOOLKIT_ROOT_DIR}" CACHE STRING "")
+  set(CAFFE2_THIRD_PARTY_ROOT "${PROJECT_SOURCE_DIR}/third_party" CACHE STRING "")
+  set(KINETO_SOURCE_DIR "${CAFFE2_THIRD_PARTY_ROOT}/kineto/libkineto" CACHE STRING "")
+  set(KINETO_BUILD_TESTS OFF CACHE BOOL "")
+  set(KINETO_LIBRARY_TYPE "static" CACHE STRING "")
+  set(CUDA_SOURCE_DIR "${CUDA_TOOLKIT_ROOT_DIR}" CACHE STRING "")
 
-    message(STATUS "Configuring Kineto dependency:")
-    message(STATUS "  KINETO_SOURCE_DIR = ${KINETO_SOURCE_DIR}")
-    message(STATUS "  KINETO_BUILD_TESTS = ${KINETO_BUILD_TESTS}")
-    message(STATUS "  KINETO_LIBRARY_TYPE = ${KINETO_LIBRARY_TYPE}")
-    message(STATUS "  CUDA_SOURCE_DIR = ${CUDA_SOURCE_DIR}")
+  message(STATUS "Configuring Kineto dependency:")
+  message(STATUS "  KINETO_SOURCE_DIR = ${KINETO_SOURCE_DIR}")
+  message(STATUS "  KINETO_BUILD_TESTS = ${KINETO_BUILD_TESTS}")
+  message(STATUS "  KINETO_LIBRARY_TYPE = ${KINETO_LIBRARY_TYPE}")
+  message(STATUS "  CUDA_SOURCE_DIR = ${CUDA_SOURCE_DIR}")
 
-    if(EXISTS ${CUDA_SOURCE_DIR}/extras/CUPTI/include)
-      set(CUPTI_INCLUDE_DIR "${CUDA_SOURCE_DIR}/extras/CUPTI/include")
-    elseif(EXISTS ${CUDA_SOURCE_DIR}/include/cupti.h)
-      set(CUPTI_INCLUDE_DIR "${CUDA_SOURCE_DIR}/include")
+  if(EXISTS ${CUDA_SOURCE_DIR}/extras/CUPTI/lib64/libcupti_static.a)
+    set(CUDA_cupti_LIBRARY "${CUDA_SOURCE_DIR}/extras/CUPTI/lib64/libcupti_static.a")
+  elseif(EXISTS ${CUDA_SOURCE_DIR}/lib64/libcupti_static.a)
+    set(CUDA_cupti_LIBRARY "${CUDA_SOURCE_DIR}/lib64/libcupti_static.a")
+  elseif(USE_CUPTI_SO)
+    if(EXISTS ${CUDA_SOURCE_DIR}/extras/CUPTI/lib64/libcupti.so)
+      set(CUDA_cupti_LIBRARY "${CUDA_SOURCE_DIR}/extras/CUPTI/lib64/libcupti.so")
+    elseif(EXISTS ${CUDA_SOURCE_DIR}/lib64/libcupti.so)
+      set(CUDA_cupti_LIBRARY "${CUDA_SOURCE_DIR}/lib64/libcupti.so")
     endif()
-
-    if((NOT DEFINED CUDA_cupti_LIBRARY) OR (${CUDA_cupti_LIBRARY} STREQUAL "CUDA_cupti_LIBRARY-NOTFOUND"))
-      if(EXISTS ${CUDA_SOURCE_DIR}/extras/CUPTI/lib64/libcupti_static.a)
-        set(CUDA_cupti_LIBRARY "${CUDA_SOURCE_DIR}/extras/CUPTI/lib64/libcupti_static.a")
-      elseif(EXISTS ${CUDA_SOURCE_DIR}/lib64/libcupti_static.a)
-        set(CUDA_cupti_LIBRARY "${CUDA_SOURCE_DIR}/lib64/libcupti_static.a")
-      elseif(EXISTS ${CUDA_SOURCE_DIR}/extras/CUPTI/lib64/libcupti.so)
-        set(CUDA_cupti_LIBRARY "${CUDA_SOURCE_DIR}/extras/CUPTI/lib64/libcupti.so")
-      elseif(EXISTS ${CUDA_SOURCE_DIR}/lib64/libcupti.so)
-        set(CUDA_cupti_LIBRARY "${CUDA_SOURCE_DIR}/lib64/libcupti.so")
-      endif()
-    endif()
-    message(STATUS "  CUDA_cupti_LIBRARY = ${CUDA_cupti_LIBRARY}")
-    message(STATUS "  CUPTI_INCLUDE_DIR = ${CUPTI_INCLUDE_DIR}")
-
-    add_subdirectory("${KINETO_SOURCE_DIR}")
-    message(STATUS "Configured Kineto as a dependency.")
   endif()
 
-  list(APPEND Caffe2_DEPENDENCY_LIBS kineto)
+  if(EXISTS ${CUDA_SOURCE_DIR}/extras/CUPTI/include)
+    set(CUPTI_INCLUDE_DIR "${CUDA_SOURCE_DIR}/extras/CUPTI/include")
+  elseif(EXISTS ${CUDA_SOURCE_DIR}/include/cupti.h)
+    set(CUPTI_INCLUDE_DIR "${CUDA_SOURCE_DIR}/include")
+  endif()
+
+  set(FOUND_CUPTI FALSE)
+  if((DEFINED CUPTI_INCLUDE_DIR) AND (DEFINED CUDA_cupti_LIBRARY))
+    if((CUDA_cupti_LIBRARY MATCHES "libcupti_static.a") OR ((CUDA_cupti_LIBRARY MATCHES "libcupti.so") AND (USE_CUPTI_SO)))
+      set(FOUND_CUPTI TRUE)
+    endif()
+  endif()
+
+  if(FOUND_CUPTI)
+    message(STATUS "  CUDA_cupti_LIBRARY = ${CUDA_cupti_LIBRARY}")
+    message(STATUS "  CUPTI_INCLUDE_DIR = ${CUPTI_INCLUDE_DIR}")
+    if(NOT TARGET kineto)
+      add_subdirectory("${KINETO_SOURCE_DIR}")
+    endif()
+    string(APPEND CMAKE_CXX_FLAGS " -DUSE_KINETO")
+    list(APPEND Caffe2_DEPENDENCY_LIBS kineto)
+    message(STATUS "Configured Kineto")
+  else()
+    message(STATUS "Could not find CUPTI library, skipping Kineto build")
+    set(USE_KINETO OFF)
+  endif()
 endif()
+
diff --git a/setup.py b/setup.py
index 50983a89ad5..42fa68e94de 100644
--- a/setup.py
+++ b/setup.py
@@ -33,8 +33,8 @@
 #   USE_FBGEMM=0
 #     disables the FBGEMM build
 #
-#   USE_KINETO=1
-#     enables experimental usage of libkineto
+#   USE_KINETO=0
+#     disables usage of libkineto library for profiling
 #
 #   USE_NUMPY=0
 #     disables the NumPy build
diff --git a/torch/csrc/autograd/profiler_kineto.cpp b/torch/csrc/autograd/profiler_kineto.cpp
index 1c3c351eeb0..df4917b3520 100644
--- a/torch/csrc/autograd/profiler_kineto.cpp
+++ b/torch/csrc/autograd/profiler_kineto.cpp
@@ -7,6 +7,7 @@
 
 #ifdef USE_KINETO
 #include <pthread.h>
+#include <torch/cuda.h>
 #include <libkineto.h>
 #endif
 
@@ -241,7 +242,7 @@ void prepareProfiler(
   }
 
   if (!libkineto::api().isProfilerRegistered()) {
-    libkineto_init();
+    libkineto_init(!torch::cuda::is_available());
     libkineto::api().suppressLogMessages();
   }