-
|
I am attempting to run the GHZ circuit code in https://docs.nvidia.com/cuda/cuquantum/latest/appliance/cusvaer.html, specifically #!/usr/bin/env python
import sys
from qiskit import QuantumCircuit, transpile
from cusvaer.backends import StatevectorSimulator
# import mpi4py here to call MPI_Init()
from mpi4py import MPI
def create_ghz_circuit(n_qubits):
ghz = QuantumCircuit(n_qubits)
ghz.h(0)
for qubit in range(n_qubits - 1):
ghz.cx(qubit, qubit + 1)
ghz.measure_all()
return ghz
n_qubits = int(sys.argv[2])
#n_qubits = 29
print(f"mpi4py: rank: {MPI.COMM_WORLD.Get_rank()}, size: {MPI.COMM_WORLD.Get_size()}")
circuit = create_ghz_circuit(n_qubits)
simulator = StatevectorSimulator()
simulator.set_options(precision='double')
circuit = transpile(circuit, simulator)
job = simulator.run(circuit, shots=1024)
result = job.result()
print(f"Result: rank: {result.mpi_rank}, size: {result.num_mpi_processes}")
if result.mpi_rank == 0:
print(result.get_counts())With the command What could possible be wrong in this setup? Guidance is appreciated. |
Beta Was this translation helpful? Give feedback.
Replies: 2 comments 22 replies
-
|
Hi @rht, |
Beta Was this translation helpful? Give feedback.
-
|
I replaced Then I cp-ed ghz_cusvaer.py to /home/cuquantum. Got this error message |
Beta Was this translation helpful? Give feedback.
-
|
There would be some connectivity issue. Could you please share the following items?
|
Beta Was this translation helpful? Give feedback.
-
|
Thank you for the help so far! Output of /usr/local/ucx/bin/ucx_info -d |
Beta Was this translation helpful? Give feedback.
-
|
I think the right network device is not used for UCX. Limiting the network devices may resolve the issue. Could you please try with the followings:
|
Beta Was this translation helpful? Give feedback.
-
This one works! Thank you very much for the answer. /usr/local/openmpi/bin/mpirun --map-by node -x UCX_TLS=all -x UCX_NET_DEVICES=ens3 -x UCX_LOG_LEVEL=info --hostfile hosts -n
4 /home/cuquantum/conda/envs/cuquantum-23.10/bin/python ghz_cusvaer.py --nqubits 31but this errored out with The UCX log is the same with when
This one failed. |
Beta Was this translation helpful? Give feedback.
-
|
Thank you for confirmation.
I think the hostfile would require max-slots=1 for localhost to ensure that only 1 process is launched on client-host1-0. This would be a general openMPI item, and you can refer to openMPI documentations (e.g., https://www.open-mpi.org/faq/?category=running#diagnose-multi-host-problems) to see if rank mapping is expected. The simulator backend assumes that each compute node has the same number of devices: https://docs.nvidia.com/cuda/cuquantum/latest/appliance/cusvaer.html#distributed-state-vector-simulation; therefore, the simulator will work, but the performance may not be optimal on your targeting system. |
Beta Was this translation helpful? Give feedback.
-
This also works! I can do 31 qubits, signifying that more than 1 GPU are activated for the computation. Which is consistent with the documentation:
|
Beta Was this translation helpful? Give feedback.
-
|
I am getting this error while running on supercomputer. I need to check if 37 qubit simulation is possible or not: mpiexec -n 16 --bind-to none --map-by node --oversubscribe -x UCX_TLS=^cma --mca coll_hcoll_enable 0 -x OMPI_MCA_coll_hcoll_enable=0 cuquantum-benchmarks circuit --frontend qiskit --backend cusvaer --benchmark quantum_volume --nqubits 36 --precision single --ngpus 1 --cusvaer-global-index-bits 3,1 --cusvaer-p2p-device-bits 3 |
Beta Was this translation helpful? Give feedback.
I think the right network device is not used for UCX. Limiting the network devices may resolve the issue. Could you please try with the followings:
-x UCX_TLS=all -x UCX_NET_DEVICES=ens3,-x UCX_TLS=all -x UCX_NET_DEVICES=docker_gwbridge?Also, by adding
-x UCX_LOG_LEVEL=info, you will be able to see more logs for the network devices.