Subject: Help with AutoDock-GPU installation – CUDA Verification Failed

[khandokaryogesh]

Hello,

I am new to molecular docking and currently trying to install and run AutoDock-GPU on my Ubuntu system. While following the installation instructions for AutoDock-GPU, I encountered an issue when running the make DEVICE=CUDA command. Here is the error I get
Using Cuda 12.0
Compiling for targets: 52 53 60 61 62 70 72 75 80 86 87 89 90
Makefile:34: *** Cuda verification failed. Stop.

have installed CUDA 12.6, and I have verified that the necessary CUDA libraries and header files are present in the system. My GPU is an NVIDIA L40S, and nvidia-smi reports the following:

Driver Version: 560.35.03
CUDA Version: 12.6

I have set the following environment variables:

export GPU_INCLUDE_PATH=/usr/local/cuda/include
export GPU_LIBRARY_PATH=/usr/local/cuda/lib64
Despite this setup, the CUDA verification step fails, preventing the compilation of AutoDock-GPU. Could you please guide me on how to resolve this issue or identify what might be going wrong?

Any help would be greatly appreciated!

[rwxayheee]

Hi @khandokaryogesh
If you haven't changed Makefile:

AutoDock-GPU/Makefile

Lines 29 to 34 in 2e658c3

	ifeq ($(DEVICE), $(filter $(DEVICE),GPU CUDA))
	TARGETS_SUPPORTED := $(shell ./test_cuda.sh nvcc "$(GPU_INCLUDE_PATH)" "$(GPU_LIBRARY_PATH)" "$(TARGETS)" "$(DEVICE)")
	# if user specifies DEVICE=GPU the test result determines wether CUDA will be used or not
	ifeq ($(TARGETS_SUPPORTED),)
	ifeq ($(DEVICE),CUDA)
	$(error Cuda verification failed)

Based on the output you got, you were able to locate nvcc (which indicates version 12.0, not 12.6; this is likely a nvcc in your PATH but not the one you wanted) but the target numbers were not printed. It's very likely that the test executable (test_cuda) were not compiled successfully by:

AutoDock-GPU/test_cuda.sh

Line 34 in 2e658c3

$1 -I$2 -L$3 -lcuda -lcudart -o test_cuda test_cuda.cpp &> /dev/null

(which is nvcc -I$GPU_INCLUDE_PATH -L#GPU_LIBRARY_PATH -lcuda -lcudart -o test_cuda test_cuda.cpp)

I have only tested up to CUDA 12.4. I also never tested the driver version or on NVIDIA L40S. Not very sure what went wrong with the compilation of test codes, but it might be helpful to run nvcc -I$GPU_INCLUDE_PATH -L$GPU_LIBRARY_PATH -lcuda -lcudart -o test_cuda test_cuda.cpp, without &> /dev/null, to see the standard output and errors. Also, if you want to use CUDA 12.6, make sure you have the right nvcc in PATH.

[atillack]

@khandokaryogesh I agree with @rwxayheee, this is likely a case of nvcc in your path being different from the directories used. Please post the output of the test compilation nvcc -I$GPU_INCLUDE_PATH -L$GPU_LIBRARY_PATH -lcuda -lcudart -o test_cuda test_cuda.cpp.