diff --git a/.github/workflows/analyze.yaml b/.github/workflows/analyze.yaml
index 1fb61f8e1..48799df3c 100644
--- a/.github/workflows/analyze.yaml
+++ b/.github/workflows/analyze.yaml
@@ -15,7 +15,7 @@ jobs:
           cd lib && ./configure --enable-sm80 --disable-openmp && cd ..
       - name: analyze
         run: |
-          cd lib && scan-build -o ../_analyze --use-cc clang make -j 32
+          cd lib && scan-build -o ../_analyze --use-cc clang make -j 64
       - name: upload report
         run: |
           cd _analyze
diff --git a/.github/workflows/coverage.yaml b/.github/workflows/coverage.yaml
index 195021882..6c25d65f9 100644
--- a/.github/workflows/coverage.yaml
+++ b/.github/workflows/coverage.yaml
@@ -13,7 +13,7 @@ jobs:
       - name: configure & build
         run: |
           cd lib && ./configure --enable-sm80 && cd ..
-          make -j 32 -C test COVER=1 all.tests
+          make -j 64 -C test COVER=1 all.tests
       - name: tests
         run: |
           make -C test check
diff --git a/.github/workflows/cuda-int-tests.yaml b/.github/workflows/cuda-int-tests.yaml
index 48df3a331..83c9d34f1 100644
--- a/.github/workflows/cuda-int-tests.yaml
+++ b/.github/workflows/cuda-int-tests.yaml
@@ -13,7 +13,7 @@ jobs:
       - name: configure & build
         run: |
           cd lib && ./configure --enable-sm80 && cd ..
-          make -j 32 -C test/int/nnc
+          make -j 64 -C test/int/nnc
       - name: integration tests
         run: |
           make -C test/int/nnc test
diff --git a/.github/workflows/ubsan-unit-tests.yaml b/.github/workflows/ubsan-unit-tests.yaml
index b6fffe4cd..0f4bccb5f 100644
--- a/.github/workflows/ubsan-unit-tests.yaml
+++ b/.github/workflows/ubsan-unit-tests.yaml
@@ -13,11 +13,11 @@ jobs:
       - name: configure & build
         run: |
           cd lib && ./configure --enable-sm80 && cd ..
-          make -j 32 -C lib ubsan
-          make -j 32 -C bin ubsan
-          make -j 32 -C bin/nnc ubsan
-          make -j 32 -C bin/cuda ubsan
-          make -j 32 -C test ubsan
+          make -j 64 -C lib ubsan
+          make -j 64 -C bin ubsan
+          make -j 64 -C bin/nnc ubsan
+          make -j 64 -C bin/cuda ubsan
+          make -j 64 -C test ubsan
       - name: tests
         run: |
           make -C test test
diff --git a/lib/configure b/lib/configure
index b5ed48c60..224cc6d4b 100755
--- a/lib/configure
+++ b/lib/configure
@@ -4641,7 +4641,7 @@ fi
 printf "%s\n" "yes" >&6; }
 		DEFINE_MACROS="$DEFINE_MACROS-D HAVE_CUDA_SM80 "
 
-		CUDA_COMPAT_SM80_LIB="gpu/libnnc-compat-cuda-sm80.o"
+		CUDA_COMPAT_SM80_LIB="libnnc-compat-cuda-sm80.o"
 
 		NV_SM_FLAGS="-arch=sm_80 -std=c++17 --expt-relaxed-constexpr"
 
diff --git a/lib/configure.ac b/lib/configure.ac
index 28df523d6..ba49af529 100644
--- a/lib/configure.ac
+++ b/lib/configure.ac
@@ -180,7 +180,7 @@ if [[ -d "$cuda_prefix" ]]; then
 	if test "$sm80_support" = yes; then
 		AC_MSG_RESULT(yes)
 		AC_SUBST(DEFINE_MACROS, ["$DEFINE_MACROS-D HAVE_CUDA_SM80 "])
-		AC_SUBST(CUDA_COMPAT_SM80_LIB, ["gpu/libnnc-compat-cuda-sm80.o"])
+		AC_SUBST(CUDA_COMPAT_SM80_LIB, ["libnnc-compat-cuda-sm80.o"])
 		AC_SUBST(NV_SM_FLAGS, ["-arch=sm_80 -std=c++17 --expt-relaxed-constexpr"])
 	else
 		AC_MSG_RESULT(no)
diff --git a/lib/nnc/gpu/makefile b/lib/nnc/gpu/makefile
index 6143c058e..ebda64250 100644
--- a/lib/nnc/gpu/makefile
+++ b/lib/nnc/gpu/makefile
@@ -24,7 +24,7 @@ lib: libnnc-compat-cuda.o
 clean:
 	rm -f *.o
 
-libnnc-compat-cuda.o: $(SRC_OBJS)
+libnnc-compat-cuda.o: $(SRC_OBJS) $(CUDA_COMPAT_SM80_LIB)
 	ld -r $^ -o $@
 
 libnnc-compat-cuda-sm80.o: $(FLASH_ATTN_SRC_OBJS)
diff --git a/lib/nnc/makefile b/lib/nnc/makefile
index bf3b0c4b6..1074bd3c1 100644
--- a/lib/nnc/makefile
+++ b/lib/nnc/makefile
@@ -7,7 +7,7 @@ SRCS := ccv_nnc_cmd.c ccv_nnc_tensor.c ccv_nnc_tensor_io.c ccv_nnc_stream.c ccv_
 
 SRC_OBJS := $(patsubst %.c,%.o,$(SRCS))
 
-.PHONY: release all lib clean cmd libnnc.o cmd/libnnc-cmd.o gpu/libnnc-compat-cuda.o gpu/libnnc-compat-cuda-sm80.o mfa/libnnc-compat-mfa.o mps/libnnc-compat-mps.o
+.PHONY: release all lib clean cmd libnnc.o cmd/libnnc-cmd.o gpu/libnnc-compat-cuda.o mfa/libnnc-compat-mfa.o mps/libnnc-compat-mps.o
 
 release: all
 
@@ -20,7 +20,7 @@ lib: libnnc.o
 clean:
 	${MAKE} -C ./cmd clean ; ${MAKE} -C ./gpu clean ; ${MAKE} -C ./mfa clean ; ${MAKE} -C ./mps clean ; rm -f *.o gpu/*.o
 
-libnnc.o: $(SRC_OBJS) cmd/libnnc-cmd.o $(CUDA_COMPAT_LIB) $(CUDA_COMPAT_SM80_LIB) $(MFA_COMPAT_LIB) $(MPS_COMPAT_LIB)
+libnnc.o: $(SRC_OBJS) cmd/libnnc-cmd.o $(CUDA_COMPAT_LIB) $(MFA_COMPAT_LIB) $(MPS_COMPAT_LIB)
 	ld -r $^ -o $@
 
 %.o: %.c
@@ -38,9 +38,6 @@ cmd:
 gpu/libnnc-compat-cuda.o:
 	${MAKE} -C ./gpu
 
-gpu/libnnc-compat-cuda-sm80.o:
-	${MAKE} -C ./gpu libnnc-compat-cuda-sm80.o
-
 mfa/libnnc-compat-mfa.o:
 	${MAKE} -C ./mfa