cudaq-guide

name	cudaq-guide
title	Cuda Quantum
description	CUDA-Q onboarding guide for installation, test programs, GPU simulation, QPU hardware, and quantum applications.
version	1.0.0
author	CUDA-Q
tags	["cuda-quantum","quantum-computing","onboarding","getting-started","nvidia"]
tools	["Read","Glob","Grep","Bash"]
license	Apache License 2.0
compatibility	Python 3.10+, C++ 20
metadata	{"author":"CUDA-Q","tags":["cuda-quantum","quantum-computing","onboarding","getting-started","nvidia"],"languages":["python","c++"],"domain":"quantum"}

CUDA-Q Getting Started Guide

You are a CUDA-Q expert assistant. Guide the user through the CUDA-Q platform based on their $ARGUMENTS. If no argument is given, present the full onboarding menu.

Purpose

Guide users through the CUDA-Q platform: installation, writing quantum kernels, GPU-accelerated simulation, connecting to QPU hardware, and exploring built-in applications.

Prerequisites

• Python 3.10+ (for Python installation path)
• CUDA Toolkit (for GPU-accelerated targets on Linux; not required on macOS)
• NVIDIA GPU (optional; CPU-only simulation available via qpp-cpu)
• For C++ path: Linux or WSL on Windows
• For QPU access: provider-specific credentials and account

Instructions

• Invoke with /cudaq-guide [argument]
• If no argument is given, display the full onboarding menu and ask what the user wants to explore
• Pass an argument from the routing table below to jump directly to that topic
• Read local CUDA-Q documentation files to answer questions accurately

References

Section	Doc file
Install	docs/sphinx/using/install/install.rst, docs/sphinx/using/quick_start.rst
Test Program	docs/sphinx/using/basics/kernel_intro.rst, docs/sphinx/using/basics/build_kernel.rst
GPU Simulation	docs/sphinx/using/backends/sims/svsims.rst, docs/sphinx/using/examples/multi_gpu_workflows.rst
QPU	docs/sphinx/using/backends/hardware.rst, docs/sphinx/using/backends/cloud.rst
Applications	docs/sphinx/using/applications.rst
Parallelize	docs/sphinx/using/examples/multi_gpu_workflows.rst

Routing by Argument

Argument	Action
install	Walk through installation (see Install section)
test-program	Build and run a Bell state kernel to verify CUDA-Q is working properly
gpu-sim	Explain GPU-accelerated simulation targets (see GPU Simulation section)
qpu	Explain how to run on real QPU hardware (see QPU section)
applications	Showcase what can be built with CUDA-Q (see Applications section)
parallelize	Show how to run circuits in parallel across multiple QPUs (see Parallelize section)
(none)	Print the full menu below and ask what they'd like to explore

---

Full Menu (no argument)

Present this when invoked with no argument

CUDA-Q Getting Started

CUDA-Q is NVIDIA's unified quantum-classical programming model for CPUs, GPUs, and QPUs.
Supports Python and C++. Docs https://nvidia.github.io/cuda-quantum/

Choose a topic
  /cudaq-guide install         Install CUDA-Q (Python pip or C++ binary)
  /cudaq-guide test-program    Write and run your quantum kernel
  /cudaq-guide gpu-sim         Accelerate simulation on NVIDIA GPUs
  /cudaq-guide qpu             Connect to real QPU hardware
  /cudaq-guide applications    Explore what you can build
  /cudaq-guide parallelize     Run circuits in parallel across multiple QPUs

Specialized skills
  /cudaq-qec        Quantum Error Correction memory experiments
  /cudaq-chemistry  Quantum chemistry (VQE, ADAPT-VQE)
  /cudaq-add-backend  Add a new hardware backend
  /cudaq-compiler   Work with the CUDA-Q compiler IR
  /cudaq-benchmark  Benchmark and optimize performance

---

Install

Instructions

• Default to Python installation unless the user explicitly mentions C++ or the nvq++ compiler.
• After installation, always guide the user through the validation step (run the Bell state example and confirm output shows { 00:~500 11:~500 }).
• Default to GPU-accelerated targets (nvidia) unless: the user is on macOS/Apple Silicon, mentions no GPU available, or explicitly asks for CPU-only simulation - in those cases use qpp-cpu.
• Do not suggest cloud trial or Launchpad options unless the user has no local environment or asks about cloud access.

Platform notes

• Linux (x86_64, ARM64): full GPU support - pip install cudaq + CUDA Toolkit

• macOS (ARM64/Apple Silicon): CPU simulation only - pip install cudaq (no CUDA Toolkit needed)

• Windows: use WSL, then follow Linux instructions

• C++ (no sudo): bash install_cuda_quantum*.$(uname -m) --accept -- --installpath $HOME/.cudaq

• Brev (cloud, no local setup): Log in at the NVIDIA Application Hub, open a CUDA-Q workspace, then SSH in with the Brev CLI:

  brev open ${WORKSPACE_NAME}
  

  CUDA-Q and the CUDA Toolkit are pre-installed.

---

Test Program

Key concepts to explain

• @cudaq.kernel / __qpu__ marks a quantum kernel - compiled to Quake MLIR
• cudaq.qvector(N) allocates N qubits in |0⟩
• cudaq.sample() - kernel measures qubits; returns bitstring histogram (SampleResult)
• cudaq.run() - kernel returns a classical value; runs shots_count times and returns a list of those return values
• cudaq.observe() - computes expectation value ⟨H⟩ for a spin operator
• cudaq.get_state() - returns the full statevector (simulator only)

Kernel restrictions

• Only a restricted Python subset is valid inside a kernel - it compiles to Quake MLIR, not regular Python.
• NumPy and SciPy cannot be used inside a kernel. Use them outside the kernel for classical pre/post-processing.
• Kernels can call other kernels; the callee must also be a @cudaq.kernel.

For compiler internals (inspect module -> ast_bridge.py -> Quake MLIR -> QIR -> JIT), route to /cudaq-compiler.

---

GPU Simulation

To recommend the best simulation backend for the user, consult the full comparison table at https://nvidia.github.io/cuda-quantum/latest/using/backends/simulators.html

Available GPU Targets

Target	Description	Use when
nvidia (default)	Single-GPU state vector via cuStateVec (up to ~30 qubits)	Default choice for most simulations on a single GPU
nvidia --target-option fp64	Double-precision single GPU	Higher numerical precision needed (e.g. chemistry, sensitive observables)
nvidia --target-option fp32	Single-precision single GPU	Large circuits where fp64 would OOM; ~2× more qubits fit in VRAM
nvidia --target-option mgpu	Multi-GPU, pools memory across GPUs (>30 qubits)	Circuit exceeds single-GPU memory; requires MPI; use fp32 option to maximise qubit count
nvidia --target-option mqpu	Multi-QPU, one virtual QPU per GPU, parallel execution	Running many independent circuits in parallel (e.g. parameter sweeps, VQE gradients)
tensornet	Tensor network simulator	Shallow or low-entanglement circuits; qubit count exceeds statevector feasibility
qpp-cpu	CPU-only fallback (OpenMP)	No GPU available; macOS; small circuits for testing

---

QPU

When the user invokes this section, do not dump all providers at once. Instead, follow this two-step dialogue:

Step 1 - ask which technology they want

Which QPU technology are you targeting?
  1. Ion trap       (IonQ, Quantinuum)
  2. Superconducting (IQM, OQC, Anyon, TII, QCI)
  3. Neutral atom   (QuEra, Infleqtion, Pasqal)
  4. Cloud / multi-platform (AWS Braket, Scaleway)

Step 2 - once they pick a technology, ask which provider, then read the corresponding doc file and walk the user through it step by step.

Technology	Provider	Doc file
Ion trap	IonQ	docs/sphinx/using/backends/hardware/iontrap.rst (IonQ section)
Ion trap	Quantinuum	docs/sphinx/using/backends/hardware/iontrap.rst (Quantinuum section)
Superconducting	IQM	docs/sphinx/using/backends/hardware/superconducting.rst (IQM section)
Superconducting	OQC	docs/sphinx/using/backends/hardware/superconducting.rst (OQC section)
Superconducting	Anyon	docs/sphinx/using/backends/hardware/superconducting.rst (Anyon section)
Superconducting	TII	docs/sphinx/using/backends/hardware/superconducting.rst (TII section)
Superconducting	QCI	docs/sphinx/using/backends/hardware/superconducting.rst (QCI section)
Neutral atom	Infleqtion	docs/sphinx/using/backends/hardware/neutralatom.rst (Infleqtion section)
Neutral atom	QuEra	docs/sphinx/using/backends/hardware/neutralatom.rst (QuEra section)
Neutral atom	Pasqal	docs/sphinx/using/backends/hardware/neutralatom.rst (Pasqal section)
Cloud	AWS Braket	docs/sphinx/using/backends/cloud/braket.rst
Cloud	Scaleway	docs/sphinx/using/backends/cloud/scaleway.rst

After walking through the provider steps, always close with

• Test locally first with emulate=True before submitting to real hardware.
• Use cudaq.sample_async() / cudaq.observe_async() for non-blocking submission.

---

Applications

CUDA-Q ships with ready-to-run application notebooks

Category	Examples
Optimization	QAOA, ADAPT-QAOA, MaxCut
Chemistry	VQE, UCCSD, ADAPT-VQE -> see /cudaq-chemistry
Error Correction	Surface codes, QEC memory -> see /cudaq-qec
Algorithms	Grover's, Shor's, QFT, Deutsch-Jozsa, HHL
ML	Quantum neural networks, kernel methods
Simulation	Hamiltonian dynamics, Trotter evolution
Finance	Portfolio optimization, Monte Carlo

Point to sub-skills for specialized topics

• /cudaq-qec - full QEC memory experiment walkthrough
• /cudaq-chemistry - VQE and ADAPT-VQE for molecular energies
• /cudaq-benchmark - performance profiling and multi-GPU scaling

---

Parallelize

CUDA-Q supports three distinct multi-GPU parallelization strategies — pick based on what you are trying to scale.

Goal	Strategy	API / Target
Single circuit too large for one GPU	Pool GPU memory across GPUs	nvidia + option='mgpu,fp32'
Many independent circuits at once	Run circuits in parallel	nvidia + option='mqpu'
Many independent shots of one circuit	Split shots across MPI ranks	mpi4py + nvidia (one rank per GPU)
Large Hamiltonian expectation value	Distribute terms across GPUs	mqpu + execution=cudaq.parallel.thread

Strategy 1 — mgpu: statevector sharding (large circuits)

The mgpu target has cuStateVec split the statevector across N GPUs via GPU-aware MPI. This extends the reachable qubit count:

GPUs	fp32 limit
1	~30 qubits (~32 GB on H100 80 GB)
2	~31 qubits
4	~32 qubits
8	~33–34 qubits (~69 GB/GPU → n=36 fits 8×H100)

import cudaq
cudaq.set_target('nvidia', option='mgpu,fp32')   # fp32 halves memory vs fp64

@cudaq.kernel
def large_circuit(n: int):
    q = cudaq.qvector(n)
    h(q[0])
    for i in range(n - 1):
        cx(q[i], q[i + 1])
    mz(q)

result = cudaq.sample(large_circuit, 34, shots_count=256)
cudaq.mpi.finalize()

Launch with MPI (one rank per GPU):

mpirun -n 4 python your_script.py
# or via Slurm:
srun -N 1 --ntasks=4 --gpus-per-node=4 python your_script.py

Noise models work on the mgpu target via trajectory simulation, but each shot runs sequentially — so mgpu noisy is slower than mpi4py shot-splitting for practical shot counts (see benchmark below).

Cray MPICH / HPC systems: GTL preload required

On Cray systems (e.g. NREL Kestrel), cuStateVec's SVSwapWorkerExecute passes GPU buffer pointers directly to MPI_Waitall. Cray MPICH's default SHM transport calls process_vm_readv on those GPU pointers → SIGABRT.

Fix: enable GPU-aware MPI via the Cray GTL library. These env vars must be set before Python starts (cannot be os.environ inside the script). Use a wrapper script:

#!/bin/bash
# run_mgpu.sh
export MPICH_GPU_SUPPORT_ENABLED=1
export LD_LIBRARY_PATH=/nopt/cuda/12.4/lib64${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}
export LD_PRELOAD=/opt/cray/pe/mpich/8.1.28/gtl/lib/libmpi_gtl_cuda.so

# Also point cuStateVec to the correct libmpi.so (Cray names it differently)
export CUDAQ_MGPU_LIB_MPI=/opt/cray/pe/mpich/8.1.28/ofi/gnu/10.3/lib/libmpi.so
export CUDAQ_MGPU_COMM_PLUGIN_TYPE=MPICH

exec /path/to/python "$@"

Then launch:

srun -N 2 --ntasks-per-node=4 --gpus-per-node=4 ./run_mgpu.sh your_script.py

Without LD_PRELOAD, you will see:

process_vm_readv: Bad address
Assertion failed in .../cray_common_memops.c at line 461: 0
MPICH ERROR [Rank 0] - Abort(1): Internal error

Strategy 2 — mqpu: parallel circuit execution

The mqpu option maps one virtual QPU to each GPU. Dispatch circuits asynchronously with qpu_id to run them simultaneously.

import cudaq

cudaq.set_target('nvidia', option='mqpu')
n_qpus = cudaq.get_platform().num_qpus()

futures = [
    cudaq.sample_async(kernel, *params, shots_count=shots // n_qpus,
                       qpu_id=i % n_qpus)
    for i, params in enumerate(param_sets)
]
results = [f.get() for f in futures]

For shot-splitting a single circuit across N GPUs:

# Submit N_SHOTS/n_qpus shots to each GPU, then merge
futures = [cudaq.sample_async(kernel, *args,
               shots_count=N_SHOTS // n_qpus, qpu_id=i)
           for i in range(n_qpus)]
combined = sum((f.get() for f in futures), cudaq.SampleResult())

Strategy 3 — mpi4py shot-splitting (noisy, multi-node)

For noisy trajectory simulation with many shots, the fastest strategy is to split shots across independent MPI ranks, each owning one GPU. This is true parallel execution — each rank runs its fraction of shots independently.

# Each MPI rank owns one GPU
import os
from mpi4py import MPI
import cudaq

comm = MPI.COMM_WORLD
rank, size = comm.rank, comm.size

# Bind GPU before any CUDA import
os.environ['CUDA_VISIBLE_DEVICES'] = os.environ.get('SLURM_LOCALID', str(rank))
cudaq.set_target('nvidia')   # single GPU per rank

my_shots = N_SHOTS // size   # divide shots evenly
counts_local = cudaq.sample(kernel, *args, shots_count=my_shots,
                             noise_model=noise_model)

# Gather and merge on rank 0
all_counts = comm.gather(counts_local, root=0)
if rank == 0:
    merged = sum(all_counts, {})

Launch with one rank per GPU:

srun -N 2 --ntasks-per-node=2 --gpus-per-node=2 python your_script.py

Strategy comparison: benchmark (DQA, 28–36 qubits, 256 shots, H100 80 GB)

Noisy trajectory simulation (p₁=0.0001, p₂=0.001, TIMESTEPS=20):

n_y	Qubits	Strategy	GPUs	Time (s)	Speedup
14	28	1-GPU baseline	1	808	1×
14	28	mqpu (shot-split)	2	283	2.85×
14	28	mpi4py (shot-split)	4	144	5.60×
14	28	mgpu (statevector)	8	198	4.07×
16	32	1-GPU	1	OOM	—
16	32	mgpu noiseless	8	12.6	—
16	32	mgpu noisy	8	>1764 (cancelled)	—
17	34	mgpu noiseless	8	33.5	—
18	36	mgpu noiseless	8	122	—

Key findings:

• For noisy simulation at n_y ≤ 14 (single H100 fits): mpi4py shot-splitting is fastest — true parallel shot execution scales near-linearly
• mgpu noisy is slower than mpi4py at the same n_y despite more GPUs: trajectory shots run sequentially, only memory is distributed
• For noiseless simulation at n_y ≥ 16 (single H100 OOM): mgpu is the only viable approach — extends reach from ~30 to ~34 qubits on 8 H100s
• n_y=16 noisy fits in memory on 8-GPU mgpu (4.3 GB/GPU) but each call takes >30 min — impractical

Hamiltonian batching

For a single kernel with a large Hamiltonian, add execution= to cudaq.observe — no other code change needed.

# Single node, multiple GPUs
result = cudaq.observe(kernel, hamiltonian, *args,
                       execution=cudaq.parallel.thread)

# Multi-node via MPI
result = cudaq.observe(kernel, hamiltonian, *args,
                       execution=cudaq.parallel.mpi)

See the docs above for complete working examples of all patterns.

---

Porting Circuits from Another Framework to CUDA-Q

When translating circuits or gates from any framework (Qiskit, Cirq, PennyLane, …) to CUDA-Q, silent correctness errors are common. The circuit compiles and runs without error but produces wrong results — often dismissed as numerical noise or shot noise. The same two checks apply regardless of the source framework.

Rule 1: validate every new gate with a unitary test

Before integrating any ported or hand-decomposed gate into a larger circuit, compare its unitary matrix against the reference from the source framework. See the ready-to-use template: templates/gate_validation.py in this skill folder.

Key points:

• Use cudaq.get_unitary(kernel, *params) with qpp-cpu — no GPU needed.
• Compare absolute values |U_cudaq| vs |U_ref| to ignore global phase.
• Use a non-trivial parameter value (e.g. beta=0.5, not 0 or π); trivial values can mask missing phase terms.
• Run this test for every non-trivial gate before using it in a circuit. Errors compound over layers and look like noise at the circuit level.

Common failure mode: gates defined by a product of Pauli exponentials (parametric swap-family, XY, iSWAP gates) often carry an odd-parity phase that naive Rxx+Ryy decompositions omit. Always derive the decomposition from the full unitary definition, not from a partial circuit identity. The unitary test will catch this immediately.

Rule 2: verify bit ordering for any new framework

Every quantum framework has its own qubit-to-bitstring convention. Mismatched indexing causes scrambled register assignments that are hard to detect without an explicit test. The template templates/gate_validation.py includes a bit-ordering check (prepare a known basis state, assert the correct bitstring).

Run it once when starting work with CUDA-Q or when upgrading to a new version. For reference, CUDA-Q uses big-endian (qubit 0 = leftmost/MSB in bitstrings), while many other frameworks use little-endian.

Rule 3: write kernels generically from the start

Write CUDA-Q kernels to be parameterized over system size (number of qubits, layers, register partitions) rather than hard-coding those values. Retrofitting general kernels later is error-prone and time-consuming.

# Prefer: parameterized over n_qubits
@cudaq.kernel
def ansatz(params: list[float], n_qubits: int, n_layers: int):
    q = cudaq.qvector(n_qubits)
    for layer in range(n_layers):
        for i in range(n_qubits):
            ry(params[layer * n_qubits + i], q[i])
        for i in range(n_qubits - 1):
            cx(q[i], q[i + 1])

# Avoid: hard-coded size
@cudaq.kernel
def ansatz_4q(params: list[float]):
    q = cudaq.qvector(4)   # must rewrite for every system size
    ...

Porting checklist

• Run templates/gate_validation.py for every new or ported gate
• Include a non-trivial parameter value in the unitary test
• Run the bit-ordering check once per framework / CUDA-Q version upgrade
• Verify full-circuit output against the source framework for a small system (2–4 qubits) before scaling up
• Write kernels parameterized over system size from the start

---

Noisy Circuit Simulation

Choosing the right evaluation method

Method	API	Speed	Noise support	Notes
Statevector loop	cudaq.get_state() + Python loop	Slow — iterates 2ⁿ states in Python	No	Avoid for n ≥ 8
Pauli observe (density matrix)	cudaq.observe() with noise model	Exact under noise	Yes	Requires nvidia-mgpu or dm target
Shot-based observe	cudaq.observe() with shots_count	Medium	Yes	Works on any noisy target
Shot sampling	cudaq.sample() + post-processing	Medium	Yes	Needed when penalties cannot be expressed as Pauli terms

Using cudaq.observe() with noise

cudaq.observe() supports noise in two ways:

1. Density-matrix target — set cudaq.set_target("dm") (or nvidia-mgpu with noise); cudaq.observe() then computes the exact noisy expectation value via the density matrix.
2. Shot-based — pass shots_count=N to cudaq.observe(); the kernel is sampled N times under the noise model and the expectation value is estimated from those shots.

# Shot-based noisy observe
result = cudaq.observe(kernel, hamiltonian, *params,
                       noise_model=noise_model, shots_count=N_SHOTS)
expectation = result.expectation()

Encoding constraint penalties in the Hamiltonian

If your cost function includes constraint penalties (e.g. Hamming-weight penalties, feasibility constraints), these can often be encoded directly in the spin Hamiltonian as additional Pauli terms, allowing cudaq.observe() to be used for noisy evaluation without any post-processing:

# Example: add a penalty λ·(Σ Zᵢ - k)² to the cost Hamiltonian
# This is exact for quadratic constraints; higher-order constraints
# require more Pauli terms but are still expressible.
cost_ham = objective_hamiltonian + penalty_weight * constraint_hamiltonian
result = cudaq.observe(kernel, cost_ham, *params, noise_model=noise_model,
                       shots_count=N_SHOTS)

When encoding into the Hamiltonian is not straightforward (e.g. non-linear penalties, penalties involving classical post-selection), use shot sampling

• manual post-processing instead:

counts = cudaq.sample(kernel, *params, noise_model=noise_model,
                      shots_count=N_SHOTS)
cost_noisy = sum(full_cost(bs) * cnt / N_SHOTS for bs, cnt in counts.items())

In either case, ensure the noisy and noiseless evaluations use the same cost function (including all penalty terms). A missing penalty term causes off-constraint bitstrings to appear cheap under noise, producing cost_noisy < cost_ideal — a physically wrong result.

Noise model parameters

Near-term device realistic values:

• p1 ≈ 1e-4 (0.01%) — single-qubit gate depolarizing rate
• p2 ≈ 1e-3 (0.1%) — two-qubit gate depolarizing rate

Using p2 = 0.01 (1%) will produce noise levels that dominate the signal entirely for circuits with more than ~100 two-qubit gates.

Noise study across system sizes: use absolute metrics and fixed circuit depth

When benchmarking noise impact across different system sizes:

1. Fix circuit depth (number of layers / timesteps) to the same value for all sizes. If depth scales with system size, small systems have too few layers for the ansatz to converge — lack of expressivity dominates over noise and masks the noise signal, making the comparison meaningless.

2. Use absolute metrics: report Δcost = cost_noisy − cost_ideal (absolute noise-induced increase). Avoid relative metrics such as Δcost / cost_ideal — the denominator grows with system size, making noise appear to decrease even when the absolute impact is increasing.

3. Visualization: keep all three panels of a noise comparison figure in the same absolute cost units so they tell a consistent story. Panels using different denominators can visually contradict each other and are misleading.

---

Limitations

• GPU simulation requires Linux (x86_64 or ARM64); macOS is CPU-only
• Multi-GPU mgpu target requires MPI
• Kernel code must use a restricted Python subset; NumPy/SciPy are not allowed inside kernels
• QPU access requires provider-specific credentials and accounts

Troubleshooting

• Import error after pip install cudaq: Ensure Python 3.10+ and a supported OS (Linux or macOS)
• No GPU detected: Verify CUDA Toolkit is installed and nvidia-smi shows your GPU; fall back to qpp-cpu
• Kernel compile error: Check that only supported Python constructs are used inside @cudaq.kernel
• QPU submission fails: Confirm credentials are set as environment variables per the provider docs