rtx2080ti Runtime API error 73

[wangniancai]

I try to run bundlefusion with nvidia rtx2080 ti ,cude error ,Source\SiftGPU\SiftPyramid.cpp(387) : cudaSafeCall() Runtime API error 73: an illegal instruction was encountered.
1080& 1080ti worked

[ll-z]

I try to run bundlefusion with nvidia rtx2080S and the ProgramCU.cu file didn't work. Have you solved your problem？

[FishHe]

The code can compile, however can not work on 2080. Hope for a solution.

[FishHe]

When I use GTX2080, the program stops at
unsigned int ProgramCU::ReshapeFeatureList(CuTexImage* raw, CuTexImage* out, int* d_featureCount, float keyLocScale)
and i use printf to test ReshapeFeatureList:

	unsigned int res;
        printf("Test begin!");
	cudaMemcpy(&res, d_featureCount, sizeof(int), cudaMemcpyDeviceToHost);
        printf("Test pass!");

cmd will only print "Test begin!", and will stops at cudaMemcpy().
I don't know how to solve the bug.

[stotko]

I am also observing the mentioned GPU hangs. I submitted a respective pull request (see #56) to fix this issue. Until it gets eventually merged, you can apply and try the changes locally on your machines.

[FishHe]

Thanks! @stotko
It (#56) really works!

[zhoushengdi]

@wangniancai Hi,I came across the same error, and I have fixed it in my case. This error is from the CMakeLists configuration, cause I run it in RTX 2070, but the default config for"nvcc -arch " is sm_61, which corresponds to the GTX 1080, so I just modify the "set(CUDA_NVCC_FLAGS -arch=sm_61;-03;-G;-g;-std=c++11)" to "set(CUDA_NVCC_FLAGS -arch=sm_75;-03;-G;-g;-std=c++11)" in CMakeLists.txt(around row 55). And finally, it worked.

I try to run bundlefusion with nvidia rtx2080 ti ,cude error ,Source\SiftGPU\SiftPyramid.cpp(387) : cudaSafeCall() Runtime API error 73: an illegal instruction was encountered. 1080& 1080ti worked

[busybeaver42]

Running on notebook with RTX 2070 doesn't work.
I have check #56. and it is included and I have changed the CMakeList.txt file:
CUDA_NVCC_FLAGS -arch=sm_75;-03;-G;-g;-std=c++11)
I try to run bundlefusion with nvidia rtx2070 ,cude error ,Source\SiftGPU\SiftPyramid.cpp(387) : cudaSafeCall() Runtime

Notebook CPU Kerne GHZ max GHz Display
SCHENKER XMG PRO 15 – M19gfp Gen9 i7-9750H 6/12 2,60 GHz 4,50 GHz 15,6",FHD, 144 Hz, IPS

NVIDIA Geforce (RTX) 2070 : nvidia-smi [8GB - cuda cores: 2304]
Compute capability = 7.5 ((Turing))//see https://developer.nvidia.com/cuda-gpus AND https://en.wikipedia.org/wiki/CUDA
sm_75

First error: terminate called after throwing an instance of 'ml::MLibException'
what(): process:46: an illegal memory access was encountered:process
if I start and start again:
/media/hd2/Lab/SLAM/BundleFusion_Ubuntu_Pangolin/src/SiftGPU/SiftPyramid.cpp(387) : cudaSafeCall() Runtime API error 700: an illegal memory access was encountered.

[Z-Jeff]

Running on notebook with RTX 2070 doesn't work. I have check #56. and it is included and I have changed the CMakeList.txt file: CUDA_NVCC_FLAGS -arch=sm_75;-03;-G;-g;-std=c++11) I try to run bundlefusion with nvidia rtx2070 ,cude error ,Source\SiftGPU\SiftPyramid.cpp(387) : cudaSafeCall() Runtime

Notebook CPU Kerne GHZ max GHz Display SCHENKER XMG PRO 15 – M19gfp Gen9 i7-9750H 6/12 2,60 GHz 4,50 GHz 15,6",FHD, 144 Hz, IPS

NVIDIA Geforce (RTX) 2070 : nvidia-smi [8GB - cuda cores: 2304] Compute capability = 7.5 ((Turing))//see https://developer.nvidia.com/cuda-gpus AND https://en.wikipedia.org/wiki/CUDA sm_75

First error: terminate called after throwing an instance of 'ml::MLibException' what(): process:46: an illegal memory access was encountered:process if I start and start again: /media/hd2/Lab/SLAM/BundleFusion_Ubuntu_Pangolin/src/SiftGPU/SiftPyramid.cpp(387) : cudaSafeCall() Runtime API error 700: an illegal memory access was encountered.

My laptop's GPU is GTX1050, I have checked CUDA_NVCC_FLAGS in CMakeLists.txt, but there still occurs an error:

terminate called after throwing an instance of 'ml::MLibException'
  what():  process:46: an illegal memory access was encountered:process

I think the problem is same with yours. Have you solved this?

[busybeaver42]

No, is still not working. Perhaps my NVIDIA Geforce (RTX) 2070 : nvidia-smi [8GB - cuda cores: 2304] has to less memory or BundleFusion need to much. But I don't see a possibility to increase my memory. On the other hand I don't see a way to reduce the memory consumption from BundleFusion. So I can't do anything. I have reached a blind alley.

[sysu17363087]

No, is still not working. Perhaps my NVIDIA Geforce (RTX) 2070 : nvidia-smi [8GB - cuda cores: 2304] has to less memory or BundleFusion need to much. But I don't see a possibility to increase my memory. On the other hand I don't see a way to reduce the memory consumption from BundleFusion. So I can't do anything. I have reached a blind alley.

hi,did you fix this problem at last? I met the same problem:(

[LSK0821]

I met the same problem, I use RTX 30 with CUDA11.
/home/liusikang/Code/BundleFusion_Ubuntu_V0/FriedLiver/Source/SiftGPU/SiftPyramid.cpp(387) : cudaSafeCall() Runtime API error 715: an illegal instruction was encountered. Killed