Cuda 9.1/9.2 support

[moskalenko]

Hi,

Will ASE_ANI work with cuda newer than 9.0? If not, would it be possible to have it updated to support current CUDA release(s)?

Thanks,

Alex

[isayev]

Hey, @moskalenko it should work with CUDA 9.1/9.2 without any troubles. We are routinely run code with 9.1. We will update next potential with the latest CUDA release too.

[moskalenko]

Olexandr, Thank you for the reply. There are shared objects in ASE_ANI/lib/ that are explicitly linked against cuda-9.0 and there is no documentation as to what is needed to build at least some of them. This makes it difficult to verify the above claim.

[Jussmith01]

Hi,

You are correct, the binaries we have built will only work with CUDA9.0. The next release of the binaries will include upgrades to the latest CUDA (this should be sometime this summer). If you have a need to run it on a system with CUDA > 9.0 then binaries will have to be built. However, currently the source code is not publicly available and we do not have a time frame on when it will become available. Contact Adrian Roitberg at [email protected] if you would like to discuss installing this in a specific place/server.

Best,
Justin

[moskalenko]

Adrian emailed me last night, so I'll follow-up about the source code with him. Fortunately, we're both at the same University and know each other.

[Jussmith01]

Alex,

A quick update,

We now have a CUDA 9.2 version in the branch 'test_branch'. We have a few people running tests on this branch to ensure a smooth transition to our newest model. If you'd like, you can try to install this branch. Let me know if you have any trouble and I will be happy to help. Also, I will be in Gainesville next week if you need any assistance.

Best,
Justin

[moskalenko]

Thanks, we'll try it out soon!

[moskalenko]

Testing under cuda/9.2, which we'll be deploying on the cluster relatively soon

$ python3 ani_quicktest.py
van Der Waals correction will be unavailable. Please install ased3
Initial Energy: -2078.5028228165684
Optimizing...
Step Time Energy fmax
LBFGS: 0 11:48:02 -2078.502823 0.3648
LBFGS: 1 11:48:02 -2078.503865 0.1675
LBFGS: 2 11:48:02 -2078.504142 0.0507
LBFGS: 3 11:48:02 -2078.504212 0.0296
LBFGS: 4 11:48:02 -2078.504264 0.0128
LBFGS: 5 11:48:02 -2078.504266 0.0017
LBFGS: 6 11:48:02 -2078.504266 0.0001
[ANI Optimization - Total time: 0.03161001205444336 seconds]
Final Energy: -2078.50426605258

bam!

Thanks!

Does that look right?

[isayev]

Looks good to me:)

[UnixJunkie]

this issue could be closed