Getting `CUDA error: no CUDA-capable device is detected` for TeraChem energy calculation

[Contextualist]

ChemCloud version: 0.8.2

input.toml:

calctype = "energy"

[extras]

[files]

[keywords]

[model]
method = "B3LYP"
basis = "6-31g"

[molecule]
symbols = [ "H", "H",]
geometry = [ [ 0.0, 0.0, 0.0,], [ 1.3889487024901424, 0.0, 0.0,],]
charge = 0
multiplicity = 1
connectivity = []

[molecule.extras]

[molecule.identifiers.extras]

Calculation setup:

from chemcloud import CCClient
from qcio import Molecule, ProgramInput
from ase.units import Bohr

client = CCClient()
mol = Molecule(symbols=['H', 'H'], geometry=[0., 0., 0., 0.735/Bohr, 0., 0.])
inp = ProgramInput(
    molecule=mol,
    model={"method": "B3LYP", "basis": "6-31g"},
    calctype="energy",
    keywords={},
)
future_result = client.compute("terachem", inp)
output = future_result.get()

Truncated TeraChem log from output.stdout

```txt | Startfile from command line: tc.in | *********************************************************** | * TeraChem v1.9-2023.09-dev * | * Development Version * | * Chemistry at the Speed of Graphics! * | *********************************************************** | * This program may only be used in connection with * | * a valid license from PetaChem, LLC. Use of this program * | * or results thereof indicates acceptance of all terms * | * and conditions stated in the license and that a valid * | * license agreement between the user and PetaChem, LLC * | * exists. PetaChem, LLC does not warrant the correctness * | * of results or their suitability for any purpose. * | * Please email bugs, suggestions, and comments to * | * help@petachem.com * | * * | *********************************************************** | *********************************************************** | * Compiled by root Fri Sep 8 02:45:55 UTC 2023 * | * Supported architecture SMs: 52 61 70 80 89 * | * Build cuda_11.8.r11.8/compiler.31833905_0 * | * Git Version: 48f1c4b24460eb111f0d752b6dea2ab9d1eb956f * | *********************************************************** | Job started Thu Oct 12 04:36:10 2023 | On ee4a8d8c6618 (available memory: 336130 MB) | ######################################### RUNTIME INFO ########################################## | terachem tc.in | NVRM version: NVIDIA UNIX x86_64 Kernel Module 525.89.02 Wed Feb 1 23:23:25 UTC 2023 | GCC version: gcc version 11.4.0 (Ubuntu 11.4.0-1ubuntu1~22.04) [lib information truncated] | Checking Amazon Cloud instance ID... Not found | Cannot find license.dat file /opt/terachem/license.dat | TRYING THE NETWORK LICENSE... | Connecting to license server 'x.x.x.x' port 'x'... | Connected! | Checking your license... | ************************************************************** | Greetings, Martinez Group! You have x licenses in total | IN USE: x | AVAILABLE: x | ************************************************************** ----------------------------------- Processed Input file: run energy coordinates geometry.xyz charge 0 spinmult 1 method b3lyp basis 6-31g ----------------------------------- Jobname: geometry Scratch data will be generated Scratch directory: ./scr.geometry | Random number seed: 1924972868 XYZ coordinates geometry.xyz Molden File Output: ./scr.geometry/geometry.molden Using basis set: 6-31g dmrgstart not found Spin multiplicity: 1 DIIS will use up to 10 vectors. Condition number limit for the DIIS overlap matrix is 0 WF convergence threshold: 3.00e-05 Using DIIS algorithm to converge WF Maximum number of SCF iterations: 100 Incremental fock with rebuild every 8 iterations Will switch to conventional Fock if diffuse functions are detected X-matrix tolerance: 1.00e-04 PRECISION: DYNAMIC TeraChem will select linear algebra engine DFT Functional requested: b3lyp Method: B3LYP Hartree-Fock exact exchange: 0.20 Slater exchange functional: 0.80 Becke 1988 exchange functional: 0.72 Lee-Yang-Parr correlation functional: 0.81 VWN(I) correlation functional: 0.19 Wavefunction: RESTRICTED DFT grid type: 1 Using dynamic DFT grids. Initial guess generated by maximum overlap ******************************************** ***** SINGLE POINT ENERGY CALCULATIONS ***** ******************************************** Starting TensorBox... | TeraChem will select linear algebra engine CUDA error: no CUDA-capable device is detected, file tensorbox/src/tensorbox.cpp, line 35 Job terminated: Thu Oct 12 04:36:11 2023 ```

[coltonbh]

Thanks for the write-up. Fixed!

The issue was I'd poached a few GPUs for TeraChem from my debug node, but someone put python processes on underneath them and so TeraChem couldn't access the GPU. Dropped those workers and everything works now. Thanks!

[Contextualist]

Great! Thanks for the quick fix!