Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Digits 2.0 with GTX 1080 not working #925

Closed
hanyunfan opened this issue Jul 22, 2016 · 7 comments
Closed

Digits 2.0 with GTX 1080 not working #925

hanyunfan opened this issue Jul 22, 2016 · 7 comments

Comments

@hanyunfan
Copy link

Error message is:

ERROR: Check failed: status == CURAND_STATUS_SUCCES(201 VS 0)
CURAND_STATUS_LAUNCH_FAILURE

Picture attached.
20160722_094044

@IsaacYangSLA
Copy link
Contributor

@hanyunfan , can you upgrade your cuda to 8.0? The curand error is likely due to GTX 1080 and cuda 7.5 or earlier.

@hanyunfan
Copy link
Author

We are using GTX1080 driver and cuda8.0 toolkit. the default driver in CUDA8 wasn't working before. Do you suggest to use that one?

@IsaacYangSLA
Copy link
Contributor

@hanyunfan , thanks for that information. You can use newer driver than the one bundled with CUDA8. May I ask if your rebuilt your caffe after installing cuda8? You can also check if your caffe is using cuda8 libraries by typing ldd which caffe in your terminal. You should be able to see lines like libcurand.so.7.5 => /usr/local/cuda/lib64/libcurand.so.7.5 (0x00007fcc5b0a8000)

In my case, I am still using cuda 7.5, so my caffe was built with libcurand.so.7.5. Yours should have 8.0.

@hanyunfan
Copy link
Author

I see, it is still 7.5. Is there a fast way that I can rebuilt caffe with little changes.

@IsaacYangSLA
Copy link
Contributor

IsaacYangSLA commented Jul 23, 2016

If your caffe was originally built from source codes, just make sure this line CUDA_DIR := /usr/local/cuda in Makefile.config points to cuda8, and make clean && make all && make test && make runtest && make pycaffe. There are experts in Caffe's github repository who can give better answers. Mine is not the optimal approach. It just works for me.

@lukeyeager
Copy link
Member

Do you need to use 2.0 for some reason? The latest release is 4.0.
https://github.com/NVIDIA/DIGITS/releases

@hanyunfan
Copy link
Author

update. After recompile with CUDA 8, both 2.0 and 4.0 works fine.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants