diff --git a/cmake/Modules/Packages/GPU.cmake b/cmake/Modules/Packages/GPU.cmake
index 84f7f5d4c8..59bc2c065b 100644
--- a/cmake/Modules/Packages/GPU.cmake
+++ b/cmake/Modules/Packages/GPU.cmake
@@ -26,6 +26,9 @@ elseif(GPU_PREC STREQUAL "SINGLE")
   set(GPU_PREC_SETTING "SINGLE_SINGLE")
 endif()
 
+option(GPU_DEBUG "Enable debugging code of the GPU package" OFF)
+mark_as_advanced(GPU_DEBUG)
+
 file(GLOB GPU_LIB_SOURCES ${CONFIGURE_DEPENDS} ${LAMMPS_LIB_SOURCE_DIR}/gpu/[^.]*.cpp)
 file(MAKE_DIRECTORY ${LAMMPS_LIB_BINARY_DIR}/gpu)
 
@@ -85,9 +88,11 @@ if(GPU_API STREQUAL "CUDA")
   # comparison chart according to: https://en.wikipedia.org/wiki/CUDA#GPUs_supported
   if(CUDA_VERSION VERSION_LESS 8.0)
     message(FATAL_ERROR "CUDA Toolkit version 8.0 or later is required")
-  elseif(CUDA_VERSION VERSION_GREATER_EQUAL "12.0")
+  elseif(CUDA_VERSION VERSION_GREATER_EQUAL "13.0")
     message(WARNING "Untested CUDA Toolkit version ${CUDA_VERSION}. Use at your own risk")
     set(GPU_CUDA_GENCODE "-arch=all")
+  elseif(CUDA_VERSION VERSION_GREATER_EQUAL "12.0")
+    set(GPU_CUDA_GENCODE "-arch=all")
   else()
     # Kepler (GPU Arch 3.0) is supported by CUDA 5 to CUDA 10.2
     if((CUDA_VERSION VERSION_GREATER_EQUAL "5.0") AND (CUDA_VERSION VERSION_LESS "11.0"))
@@ -151,14 +156,17 @@ if(GPU_API STREQUAL "CUDA")
   add_library(gpu STATIC ${GPU_LIB_SOURCES} ${GPU_LIB_CUDPP_SOURCES} ${GPU_OBJS})
   target_link_libraries(gpu PRIVATE ${CUDA_LIBRARIES} ${CUDA_CUDA_LIBRARY})
   target_include_directories(gpu PRIVATE ${LAMMPS_LIB_BINARY_DIR}/gpu ${CUDA_INCLUDE_DIRS})
-  target_compile_definitions(gpu PRIVATE -DUSE_CUDA -D_${GPU_PREC_SETTING} -DMPI_GERYON -DUCL_NO_EXIT ${GPU_CUDA_MPS_FLAGS})
+  target_compile_definitions(gpu PRIVATE -DUSE_CUDA -D_${GPU_PREC_SETTING} ${GPU_CUDA_MPS_FLAGS})
+  if(GPU_DEBUG)
+    target_compile_definitions(gpu PRIVATE -DUCL_DEBUG -DGERYON_KERNEL_DUMP)
+  else()
+    target_compile_definitions(gpu PRIVATE -DMPI_GERYON -DUCL_NO_EXIT)
+  endif()
   if(CUDPP_OPT)
     target_include_directories(gpu PRIVATE ${LAMMPS_LIB_SOURCE_DIR}/gpu/cudpp_mini)
     target_compile_definitions(gpu PRIVATE -DUSE_CUDPP)
   endif()
 
-  target_link_libraries(lammps PRIVATE gpu)
-
   add_executable(nvc_get_devices ${LAMMPS_LIB_SOURCE_DIR}/gpu/geryon/ucl_get_devices.cpp)
   target_compile_definitions(nvc_get_devices PRIVATE -DUCL_CUDADR)
   target_link_libraries(nvc_get_devices PRIVATE ${CUDA_LIBRARIES} ${CUDA_CUDA_LIBRARY})
@@ -222,15 +230,18 @@ elseif(GPU_API STREQUAL "OPENCL")
   add_library(gpu STATIC ${GPU_LIB_SOURCES})
   target_link_libraries(gpu PRIVATE OpenCL::OpenCL)
   target_include_directories(gpu PRIVATE ${CMAKE_CURRENT_BINARY_DIR}/gpu)
-  target_compile_definitions(gpu PRIVATE -D_${GPU_PREC_SETTING} -DMPI_GERYON -DGERYON_NUMA_FISSION -DUCL_NO_EXIT)
-  target_compile_definitions(gpu PRIVATE -DUSE_OPENCL)
-
-  target_link_libraries(lammps PRIVATE gpu)
+  target_compile_definitions(gpu PRIVATE -DUSE_OPENCL -D_${GPU_PREC_SETTING})
+  if(GPU_DEBUG)
+    target_compile_definitions(gpu PRIVATE -DUCL_DEBUG -DGERYON_KERNEL_DUMP)
+  else()
+    target_compile_definitions(gpu PRIVATE -DMPI_GERYON -DGERYON_NUMA_FISSION -DUCL_NO_EXIT)
+  endif()
 
   add_executable(ocl_get_devices ${LAMMPS_LIB_SOURCE_DIR}/gpu/geryon/ucl_get_devices.cpp)
   target_compile_definitions(ocl_get_devices PRIVATE -DUCL_OPENCL)
   target_link_libraries(ocl_get_devices PRIVATE OpenCL::OpenCL)
   add_dependencies(ocl_get_devices OpenCL::OpenCL)
+
 elseif(GPU_API STREQUAL "HIP")
   if(NOT DEFINED HIP_PATH)
       if(NOT DEFINED ENV{HIP_PATH})
@@ -374,8 +385,12 @@ elseif(GPU_API STREQUAL "HIP")
 
   add_library(gpu STATIC ${GPU_LIB_SOURCES})
   target_include_directories(gpu PRIVATE ${LAMMPS_LIB_BINARY_DIR}/gpu)
-  target_compile_definitions(gpu PRIVATE -D_${GPU_PREC_SETTING} -DMPI_GERYON -DUCL_NO_EXIT)
-  target_compile_definitions(gpu PRIVATE -DUSE_HIP)
+  target_compile_definitions(gpu PRIVATE -DUSE_HIP -D_${GPU_PREC_SETTING})
+  if(GPU_DEBUG)
+    target_compile_definitions(gpu PRIVATE -DUCL_DEBUG -DGERYON_KERNEL_DUMP)
+  else()
+    target_compile_definitions(gpu PRIVATE -DMPI_GERYON -DUCL_NO_EXIT)
+  endif()
   target_link_libraries(gpu PRIVATE hip::host)
 
   if(HIP_USE_DEVICE_SORT)
@@ -433,7 +448,7 @@ elseif(GPU_API STREQUAL "HIP")
 
   add_executable(hip_get_devices ${LAMMPS_LIB_SOURCE_DIR}/gpu/geryon/ucl_get_devices.cpp)
   target_compile_definitions(hip_get_devices PRIVATE -DUCL_HIP)
-  target_link_libraries(hip_get_devices hip::host)
+  target_link_libraries(hip_get_devices PRIVATE hip::host)
 
   if(HIP_PLATFORM STREQUAL "nvcc")
     target_compile_definitions(gpu PRIVATE -D__HIP_PLATFORM_NVCC__)
@@ -458,10 +473,14 @@ elseif(GPU_API STREQUAL "HIP")
     target_compile_definitions(hip_get_devices PRIVATE -D__HIP_PLATFORM_AMD__)
     target_include_directories(hip_get_devices PRIVATE ${HIP_ROOT_DIR}/../include)
   endif()
-
-  target_link_libraries(lammps PRIVATE gpu)
 endif()
 
+if(BUILD_OMP)
+  find_package(OpenMP COMPONENTS CXX REQUIRED)
+  target_link_libraries(gpu PRIVATE OpenMP::OpenMP_CXX)
+endif()
+target_link_libraries(lammps PRIVATE gpu)
+
 set_property(GLOBAL PROPERTY "GPU_SOURCES" "${GPU_SOURCES}")
 # detect styles which have a GPU version
 RegisterStylesExt(${GPU_SOURCES_DIR} gpu GPU_SOURCES)