diff --git a/CHANGELOG b/CHANGELOG
index 4798ab4bd..da32b1b87 100644
--- a/CHANGELOG
+++ b/CHANGELOG
@@ -3,6 +3,8 @@ If not stated, FINUFFT is assumed (cuFINUFFT <=1.3 is listed separately).
 
 Master, using release name V 2.4.0 (2/11/25)
 
+* Removed FINUFFT_CUDA_ARCHITECTURES flag, as it was unnecessary duplication.
+* Enabled LTO for finufft, nvcc support is flaky at the moment.
 * Added GPU spread interp only test. Added CPU spread interp only test to cmake
 * Make attributes private in Python Plan classes and allow read-only access to
   them using properties (Andén #608).
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 6a9d93d93..8d21cfc3f 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -25,8 +25,6 @@ option(FINUFFT_BUILD_DOCS "Whether to build the FINUFFT documentation" OFF)
 # if FINUFFT_USE_DUCC0 is ON, the following options are ignored
 set(FINUFFT_FFTW_LIBRARIES "DEFAULT" CACHE STRING "Specify a custom FFTW library")
 set(FINUFFT_FFTW_SUFFIX "OpenMP" CACHE STRING "Suffix for FFTW libraries (e.g. OpenMP, Threads etc.)")
-# if FINUFFT_USE_CUDA is OFF, the following options are ignored
-set(FINUFFT_CUDA_ARCHITECTURES "native" CACHE STRING "CUDA architectures to build for (e.g. 60;70;75;)")
 # if FINUFFT_USE_CPU is OFF, the following options are ignored
 set(FINUFFT_ARCH_FLAGS "native" CACHE STRING "Compiler flags for specifying target architecture, defaults to -march=native")
 # sphinx tag (don't remove): @cmake_opts_end
@@ -77,7 +75,8 @@ set(FINUFFT_CXX_FLAGS_DEBUG
     -ggdb
     -ggdb3
     -Wall
-    -Wno-sign-compare
+    -Wextra
+    -Wpedantic
     -Wno-unknown-pragmas
 )
 
@@ -208,7 +207,12 @@ function(finufft_link_test target)
     endif()
     enable_asan(${target})
     target_compile_features(${target} PRIVATE cxx_std_17)
-    set_target_properties(${target} PROPERTIES MSVC_RUNTIME_LIBRARY "MultiThreaded$<$<CONFIG:Debug>:Debug>")
+    set_target_properties(
+        ${target}
+        PROPERTIES
+            MSVC_RUNTIME_LIBRARY "MultiThreaded$<$<CONFIG:Debug>:Debug>"
+            POSITION_INDEPENDENT_CODE ${FINUFFT_POSITION_INDEPENDENT_CODE}
+    )
     # disable deprecated warnings for tests if supported
     if(FINUFFT_HAS_NO_DEPRECATED_DECLARATIONS)
         target_compile_options(${target} PRIVATE -Wno-deprecated-declarations)
@@ -277,16 +281,14 @@ if(FINUFFT_USE_CPU)
 endif()
 
 if(FINUFFT_USE_CUDA)
-    if(NOT DEFINED FINUFFT_CUDA_ARCHITECTURES)
-        if(DEFINED CMAKE_CUDA_ARCHITECTURES)
-            set(FINUFFT_CUDA_ARCHITECTURES "{$CMAKE_CUDA_ARCHITECTURES}")
-        else()
-            message(
-                "FINUFFT WARNING: No CUDA architecture supplied via '-DFINUFFT_CUDA_ARCHITECTURES=...', defaulting to 'native'"
-            )
-            message("See: https://developer.nvidia.com/cuda-gpus for more details on what architecture to supply.")
-        endif()
+    if(NOT DEFINED CMAKE_CUDA_ARCHITECTURES)
+        message(
+            WARNING
+            "FINUFFT WARNING: No CUDA architecture supplied via '-DCMAKE_CUDA_ARCHITECTURES=...', defaulting to 'native'"
+        )
+        message(WARNING "See: https://developer.nvidia.com/cuda-gpus for more details on what architecture to supply.")
     endif()
+    set(CMAKE_CUDA_ARCHITECTURES "native")
     enable_language(CUDA)
     find_package(CUDAToolkit REQUIRED)
     add_subdirectory(src/cuda)
@@ -348,6 +350,7 @@ message(STATUS "  FINUFFT_FFTW_SUFFIX: ${FINUFFT_FFTW_SUFFIX}")
 message(STATUS "  FINUFFT_FFTW_LIBRARIES: ${FINUFFT_FFTW_LIBRARIES}")
 message(STATUS "  FINUFFT_ARCH_FLAGS: ${FINUFFT_ARCH_FLAGS}")
 message(STATUS "  FINUFFT_USE_DUCC0: ${FINUFFT_USE_DUCC0}")
+message(STATUS "  CMAKE_CUDA_ARCHITECTURES: ${CMAKE_CUDA_ARCHITECTURES}")
 # gersemi: on
 
 if(FINUFFT_ENABLE_INSTALL)
diff --git a/cmake/utils.cmake b/cmake/utils.cmake
index bcbbb5169..b5571605f 100644
--- a/cmake/utils.cmake
+++ b/cmake/utils.cmake
@@ -75,3 +75,14 @@ function(copy_dll source_target destination_target)
     unset(SOURCE_FILE)
     unset(DESTINATION_FILE)
 endfunction()
+
+include(CheckIPOSupported)
+check_ipo_supported(RESULT LTO_SUPPORTED OUTPUT LTO_ERROR)
+
+if(LTO_SUPPORTED)
+    message(STATUS "LTO is supported and enabled.")
+    set(FINUFFT_INTERPROCEDURAL_OPTIMIZATION TRUE)
+else()
+    message(WARNING "LTO is not supported: ${LTO_ERROR}")
+    set(FINUFFT_INTERPROCEDURAL_OPTIMIZATION FALSE)
+endif()
diff --git a/docs/install_gpu.rst b/docs/install_gpu.rst
index ec6a74b60..0c5b4f312 100644
--- a/docs/install_gpu.rst
+++ b/docs/install_gpu.rst
@@ -45,6 +45,8 @@ To find out your own device's compute capability without having to look it up on
 This will return a text string such as ``8.6`` which would incidate
 ``sm_86`` architecture, thus to use ``CMAKE_CUDA_ARCHITECTURES=86``.
 
+Note that by default the ``CMAKE_CUDA_ARCHITECTURES`` flag is set to ``native``, which means that the code will be compiled for the compute capability of the GPU on which the code is being compiled.
+This might not be portable so it is recommended to set this flag explicitly when building for multiple systems. A good alternative is ``all-major`` which will compile for all major compute capabilities.
 
 Testing
 -------
diff --git a/perftest/cuda/CMakeLists.txt b/perftest/cuda/CMakeLists.txt
index fdf04e723..6f440ed35 100644
--- a/perftest/cuda/CMakeLists.txt
+++ b/perftest/cuda/CMakeLists.txt
@@ -7,7 +7,7 @@ set_target_properties(
     cuperftest
     PROPERTIES
         LINKER_LANGUAGE CUDA
-        CUDA_ARCHITECTURES "${FINUFFT_CUDA_ARCHITECTURES}"
+        CUDA_ARCHITECTURES "${CMAKE_CUDA_ARCHITECTURES}"
         CUDA_STANDARD 17
         CUDA_STANDARD_REQUIRED ON
 )
diff --git a/src/cuda/CMakeLists.txt b/src/cuda/CMakeLists.txt
index eccebb17a..83eba1c96 100644
--- a/src/cuda/CMakeLists.txt
+++ b/src/cuda/CMakeLists.txt
@@ -34,18 +34,14 @@ set(FINUFFT_CUDA_FLAGS
     -fmad=true
     -restrict
     --extra-device-vectorization
-    $<$<CONFIG:Debug>:-G
-    -maxrregcount
-    64
-    >
-    >
+    -Xnvlink
+    --strip-all>
 )
 
 if(FINUFFT_SHARED_LINKING)
     add_library(cufinufft SHARED ${PRECISION_INDEPENDENT_SRC} ${PRECISION_DEPENDENT_SRC})
 else()
     add_library(cufinufft STATIC ${PRECISION_INDEPENDENT_SRC} ${PRECISION_DEPENDENT_SRC})
-    set_target_properties(cufinufft PROPERTIES POSITION_INDEPENDENT_CODE ${FINUFFT_POSITION_INDEPENDENT_CODE})
 endif()
 target_include_directories(cufinufft PUBLIC ${CUFINUFFT_INCLUDE_DIRS})
 # set target build location
@@ -54,20 +50,28 @@ set_target_properties(cufinufft PROPERTIES LIBRARY_OUTPUT_DIRECTORY "${PROJECT_B
 set_target_properties(
     cufinufft
     PROPERTIES
-        CUDA_ARCHITECTURES "${FINUFFT_CUDA_ARCHITECTURES}"
+        CUDA_ARCHITECTURES "${CMAKE_CUDA_ARCHITECTURES}"
         CUDA_SEPARABLE_COMPILATION ON
         CUDA_STANDARD 17
         CUDA_STANDARD_REQUIRED ON
         WINDOWS_EXPORT_ALL_SYMBOLS ON
         ARCHIVE_OUTPUT_DIRECTORY "${CMAKE_BINARY_DIR}"
+        INTERPROCEDURAL_OPTIMIZATION
+            OFF # LTO is not supported for CUDA for now
+        POSITION_INDEPENDENT_CODE ${FINUFFT_POSITION_INDEPENDENT_CODE}
 )
 target_compile_features(cufinufft PRIVATE cxx_std_17)
 target_compile_options(cufinufft PRIVATE ${FINUFFT_CUDA_FLAGS})
-if(WIN32 OR (BUILD_TESTING AND FINUFFT_BUILD_TESTS))
+if(WIN32 OR (BUILD_TESTING AND FINUFFT_BUILD_TESTS) OR env{CIBUILDWHEEL})
     target_link_libraries(cufinufft PUBLIC CUDA::cudart CUDA::cufft)
 else()
     target_link_libraries(cufinufft PUBLIC CUDA::cudart_static CUDA::cufft_static)
 endif()
 
+# disable deprecated warnings for tests if supported
+if(FINUFFT_HAS_NO_DEPRECATED_DECLARATIONS)
+    target_compile_options(cufinufft PRIVATE $<$<COMPILE_LANGUAGE:CUDA>:-Wno-deprecated-declarations>)
+endif()
+
 file(GLOB CUFINUFFT_PUBLIC_HEADERS "${CMAKE_SOURCE_DIR}/include/cufinufft*.h")
 set_target_properties(cufinufft PROPERTIES PUBLIC_HEADER "${CUFINUFFT_PUBLIC_HEADERS}")
diff --git a/test/CMakeLists.txt b/test/CMakeLists.txt
index 71b70ef3c..7f2184fec 100644
--- a/test/CMakeLists.txt
+++ b/test/CMakeLists.txt
@@ -40,7 +40,6 @@ if(NOT FINUFFT_USE_DUCC0)
     add_executable(fftw_lock_test fftw_lock_test.cpp)
     target_compile_features(fftw_lock_test PRIVATE cxx_std_17)
     finufft_link_test(fftw_lock_test)
-
     add_test(NAME run_fftw_lock_test COMMAND fftw_lock_test WORKING_DIRECTORY ${CMAKE_BINARY_DIR})
 endif()
 
diff --git a/test/cuda/CMakeLists.txt b/test/cuda/CMakeLists.txt
index 01de47cc4..3e1bada51 100644
--- a/test/cuda/CMakeLists.txt
+++ b/test/cuda/CMakeLists.txt
@@ -13,12 +13,12 @@ foreach(srcfile ${test_src})
     target_compile_features(${executable} PUBLIC cxx_std_17)
     set_target_properties(
         ${executable}
-        PROPERTIES LINKER_LANGUAGE CUDA CUDA_ARCHITECTURES "${FINUFFT_CUDA_ARCHITECTURES}"
+        PROPERTIES LINKER_LANGUAGE CUDA CUDA_ARCHITECTURES "${CMAKE_CUDA_ARCHITECTURES}"
     )
     message(
         STATUS
         "Adding test ${executable}"
-        " with CUDA_ARCHITECTURES=${FINUFFT_CUDA_ARCHITECTURES}"
+        " with CUDA_ARCHITECTURES=${CMAKE_CUDA_ARCHITECTURES}"
         " and INCLUDE=${CUFINUFFT_INCLUDE_DIRS}"
     )
 endforeach()