{toolchain}[dummy/dummy] goolfc v2017b#5768
Conversation
| 'https://developer.nvidia.com/compute/%(namelower)s/%(version_major_minor)s/Prod/local_installers/', | ||
| ] | ||
| sources = ['%%(namelower)s_%%(version)s_%s_linux-run' % nv_version] | ||
| #checksums = ['8496c72b16fee61889f9281449b5d633d0b358b46579175c275d85c9205fe953'] |
There was a problem hiding this comment.
@akesandgren Please fix? I have 96863423feaa50b5c1c5e1b9ec537ef7ba77576a3986652351ae43e66bcd080c as SHA256 for cuda_9.0.176_384.81_linux-run, this one you have here is the one for cuda_9.1.85_387.26_linux
There was a problem hiding this comment.
Oops, must have forgotten to send that update.
|
Test report by @boegel |
|
Test report by @boegel |
|
Going in, thanks @akesandgren! |
That's new for me (in the build of openmpi) |
|
In Not sure what is supposed to be there. |
|
Odd, build with --debug and check what the configure step says. |
|
I have this on my laptop (i.e. no infiniband for instance) |
|
@wpoely86 Have you figured this out yet? |
|
Ah, it's the UCX framework? Is that installed as a OS package? |
|
Yes, removing the rpms for UCX fixed it. It seems the configure script fails to do it correctly. |
(created using
eb --new-pr)New version of goolfc toolchain. Uses same versions of dependencies as foss/2017b but adds CUDA 9.0