GPU Access for qiskit-aer-gpu from Ubuntu VM

Question

I have NVIDIA GeForce GT740 Graphics card installed on my host machine with Windows 10 OS. I have to simulate a cluster environment in order to run qiskit codes in parallel fashion. Before actual Clustering, as POC I have created three ubuntu 20.04 VMs (one head and two compute nodes) on my host machine since qiskit-aer-gpu can be easily installed on Linux. I have used VMWare Workstation Pro for the purpose.

Now on my head node I have created a venv and installed qiskit-aer-gpu along with CUDA

• python version: 3.8.10

• qiskit-aer-gpu version: 0.11.2

• nvcc version: Cuda compilation tools, release 10.1, V10.1.243

I am running following code as an example just to verify cuda installation on my VM but it generates error

from qiskit import *
from qiskit.circuit.library import *
from qiskit.providers.aer import *
sim = AerSimulator(method= 'statevector', device='GPU')
shots = 100
depth=10
qubits = 25
circuit = transpile(QuantumVolume(qubits, depth, seed=0), backend=sim, optimization_level=0)
circuit.measure_all()
result = execute(circuit,sim,shots=shots,seed_simulator=12345).result()

Error is

Simulation failed and returned the following error message:
ERROR: Failed to load qobj: No CUDA device available!

Ubuntu VM sees host GPU like sudo lshw -C display

-display                 
   description: VGA compatible controller
   product: SVGA II Adapter
   vendor: VMware
   physical id: f
   bus info: pci@0000:00:0f.0
   version: 00
   width: 32 bits
   clock: 33MHz
   capabilities: vga_controller bus_master cap_list rom
   configuration: driver=vmwgfx latency=64
   resources: irq:16 ioport:1070(size=16) 
   memory:e8000000-efffffff memory:fe000000-fe7fffff 
   memory:c0000-dffff

I have seen some article in which VM doesnot have direct access to host GPU 1 and 2. Am I getting this no device error due to this reason?

Any help will be appreciated.

Answer 1

This might be an issue beyond qiskit-aer-gpu. To confirm or discard that, my to-go tool is PyTorch. Install pip install torch and run in a console:

from torch.cuda import is_available
is_available()

If True, submit an issue to https://github.com/Qiskit/qiskit-aer because it is, indeed, an Aer issue. If False, the problem is related to your environment and StackOverflow is probably a good place to ask.

Answer 2

i am not sure , but dont you have to install cuQuantum first ? note:i cant comment due to my reputation number

Answer 3

I have encountered exactly the same issue. The way I solved it is firstly uninstalling both qiskit_aer and qiskit_aer_gpu_cu11. Then only pip install qiskit_aer_gpu_cu11 since my CUDA version is CUDA11. Finally check it by print(simulator.available_devices()) and the result is True. Hope it works for you!