Revert cmake hackery

Author: eous

ggml/src/ggml-cuda/CMakeLists.txt

@@ -15,7 +15,6 @@ if (CUDAToolkit_FOUND)
         # 80     == Ampere, asynchronous data loading, faster tensor core instructions
         # 86     == RTX 3000, needs CUDA v11.1
         # 89     == RTX 4000, needs CUDA v11.8
-        # 120    == Blackwell, compute capability 12.8.1 minimum
         #
         # XX-virtual == compile CUDA code as PTX, do JIT compilation to binary code on first run
         # XX-real    == compile CUDA code as device code for this specific architecture
@@ -27,15 +26,15 @@ if (CUDAToolkit_FOUND)
             set(CMAKE_CUDA_ARCHITECTURES "native")
         elseif(GGML_CUDA_F16 OR GGML_CUDA_DMMV_F16)
             if (CUDAToolkit_VERSION VERSION_GREATER_EQUAL "11.8")
-                set(CMAKE_CUDA_ARCHITECTURES "60-virtual;61-virtual;70-virtual;75-virtual;80-virtual;86-real;89-real;120-real")
+                set(CMAKE_CUDA_ARCHITECTURES "60-virtual;61-virtual;70-virtual;75-virtual;80-virtual;86-real;89-real")
             else()
-                set(CMAKE_CUDA_ARCHITECTURES "60-virtual;61-virtual;70-virtual;75-virtual;80-virtual;86-real;120-real")
+                set(CMAKE_CUDA_ARCHITECTURES "60-virtual;61-virtual;70-virtual;75-virtual;80-virtual;86-real")
             endif()
         else()
             if (CUDAToolkit_VERSION VERSION_GREATER_EQUAL "11.8")
-                set(CMAKE_CUDA_ARCHITECTURES "50-virtual;61-virtual;70-virtual;75-virtual;80-virtual;86-real;89-real;120-real")
+                set(CMAKE_CUDA_ARCHITECTURES "50-virtual;61-virtual;70-virtual;75-virtual;80-virtual;86-real;89-real")
             else()
-                set(CMAKE_CUDA_ARCHITECTURES "50-virtual;61-virtual;70-virtual;75-virtual;80-virtual;86-real;120-real")
+                set(CMAKE_CUDA_ARCHITECTURES "50-virtual;61-virtual;70-virtual;75-virtual;80-virtual;86-real")
             endif()
         endif()
     endif()