Caffe doesn't work on Tesla V100 GPUs.

[BinZlP]

Issue summary

Caffe doesn't work well on Tesla V100.

This is my error message: check failed error == cudasuccess no kernel image is available for execution on the device. Stack trace:

Steps to reproduce

Install OpenPose by CMake-GUI(as installing OpenPose, CMake also build Caffe as well). Execute example codes.

Tried solutions

Give manual number for CUDA_ARCH of CMake Flag.

System configuration

• Operating system: Ubuntu 16.04
• Compiler: CUDA
• CUDA version (if applicable): 10.0
• CUDNN version (if applicable): 7.5
• BLAS:
• Python version (if using pycaffe): 3.5.1
• MATLAB version (if using matcaffe):
• GPU: Nvidia Tesla V100 32GB * 2

Issue checklist

• [x] read the guidelines and removed the first paragraph
• [x] written a short summary and detailed steps to reproduce
• [x] explained how solutions to related problems failed (tick if found none)
• [x] filled system configuration
• [x] attached relevant logs/config files (tick if not applicable)

[BinZlP]

I solved it by updating Cuda.cmake file's known_gpu_archs.

[Luckick]

Hi,

do you mean: set(Caffe_known_gpu_archs "30 35 50 52 60 61 70 75") ?

I found it is already set this way, include 70 for V100. But still fails.

Thanks.