update for cu13 builds (no ci will be provided)

CMakeLists.txt

@@ -130,7 +130,10 @@ if (LLAMA_CUBLAS)
         # 75 == int8 tensor cores
         # 80 == Ampere, asynchronous data loading, faster tensor core instructions
         message("CUDA Toolkit Version: ${CUDAToolkit_VERSION}")
-        if(CUDAToolkit_VERSION VERSION_GREATER 12)
+        if(CUDAToolkit_VERSION VERSION_GREATER_EQUAL 13)
+                add_compile_definitions(GGML_CUDA_USE_GRAPHS) #try enable cuda graphs on cu12 build
+                set(CMAKE_CUDA_ARCHITECTURES "75-virtual;80-virtual;86-virtual") # lowest CUDA 13 standard
+        elseif(CUDAToolkit_VERSION VERSION_GREATER 12)
                 add_compile_definitions(GGML_CUDA_USE_GRAPHS) #try enable cuda graphs on cu12 build
                 set(CMAKE_CUDA_ARCHITECTURES "50-virtual;61-virtual;70-virtual;75-virtual;80-virtual") # lowest CUDA 12 standard + lowest for integer intrinsics
         else()

Makefile

@@ -231,6 +231,12 @@ NVCCFLAGS += -Wno-deprecated-gpu-targets \
              -gencode arch=compute_75,code=compute_75 \
              -gencode arch=compute_80,code=compute_80
 
+else ifdef LLAMA_ARCHES_CU13
+NVCCFLAGS += -Wno-deprecated-gpu-targets \
+             -gencode arch=compute_75,code=compute_75 \
+             -gencode arch=compute_80,code=compute_80 \
+             -gencode arch=compute_86,code=compute_86
+
 else
 NVCCFLAGS += -Wno-deprecated-gpu-targets -arch=all
 endif

koboldcpp.sh

@@ -39,6 +39,9 @@ fi
 if [ -n "$ARCHES_CU12" ]; then
 	ARCHES_FLAG="LLAMA_ARCHES_CU12=1"
 fi
+if [ -n "$ARCHES_CU13" ]; then
+	ARCHES_FLAG="LLAMA_ARCHES_CU13=1"
+fi
 if [ -n "$NO_WMMA" ]; then
 	NO_WMMA_FLAG="LLAMA_NO_WMMA=1"
 fi