GPUs

The qBraid Lab GPU server is tailored for researchers and developers requiring enhanced computational capabilities. This high-performance Lab instance allows users to leverage GPUs for accelerated circuit simulation, to explore quantum machine learning applications with GPU-enabled quantum gradients, and more.

• Wide GPU access: V100, A100, H100, GH200, and B200 class GPUs, available in various configurations. Billing is in credits/minute with rates shown in your account launcher.
• Pre-configured Python environment: Activate the default environment by running qbraid envs activate default in a terminal or simply open a notebook and make sure the default kernel is selected. This environment includes GPU-optimized versions of qiskit / qiskit-aer, cudaq, pennylane / pennylane-lightning, PyTorch, TensorFlow, and JAX.

Launch GPU instance

Use the drop-down at the top of your account page to select the GPU Lab image, and then click Launch Lab.

You will first see a “Starting” panel while your server is being prepared, and once it’s ready, the status will change to “Running”. Click Launch Lab to access your GPU instance.

The GPU servers may take up to 15 minutes to launch compared to standard Lab instances, as the GPU resources are provisioned on-demand.

Available GPU Configurations

qBraid offers a variety of GPU configurations to meet different computational needs:

Instance Name	GPU Type	RAM Configuration	Credits/Min
1xA100-40GB-SXM4	1x NVIDIA A100 40GB SXM4	40GB VRAM	2.15
1xH100-80GB-PCIE	1x NVIDIA H100 80GB PCIe	80GB VRAM	4.15
2xH100-80GB-SXM5	2x NVIDIA H100 80GB SXM5	80GB VRAM each	10.63
4xH100-80GB-SXM5	4x NVIDIA H100 80GB SXM5	80GB VRAM each	20.60
8xV100-16GB	8x NVIDIA V100 16GB	16GB VRAM each	7.33
8xA100-40GB-SXM4	8x NVIDIA A100 40GB SXM4	40GB VRAM each	17.20
8xA100-80GB-SXM4	8x NVIDIA A100 80GB SXM4	80GB VRAM each	23.87
8xH100-80GB-SXM5	8x NVIDIA H100 80GB SXM5	80GB VRAM each	39.87

Further information can be retrieved using the NVIDIA System Management Interface (nvidia-smi) and NVIDIA CUDA Toolkit (nvcc) command line utilities.

Additional GPU configurations are available. Visit your account page to see the full list of options.

GPU-enabled environments

The GPU Lab image comes pre-configured with the NVIDIA cuQuantum SDK GPU simulator library, and includes GPU integrations with other popular quantum softwares packages such as Pennylane-Lightning, Qiskit Aer, and Qsim-Cirq.

Pennylane-Lightning

PennyLane is a cross-platform Python library for quantum machine learning, automatic differentiation, and optimization of hybrid quantum-classical computations. The PennyLane-Lightning-GPU plugin extends the Pennylane-Lightning state-vector simulator written in C++, and offloads to the NVIDIA cuQuantum SDK for GPU accelerated circuit simulation.

The lightning.gpu device is an extension of PennyLane’s built-in lightning.qubit device. It extends the CPU-focused Lightning simulator to run using the NVIDIA cuQuantum SDK, enabling GPU-accelerated simulation of quantum state-vector evolution.

A lightning.gpu device can be loaded using:

import pennylane as qml

dev = qml.device("lightning.qubit", wires=2)

The above device will allow all operations to be performed on the pre-configured CUDA capable GPU. If not used inside the qBraid GPU instance, or if the cuQuantum libraries are not installed in the given environment, the device will fall-back to lightning.qubit and perform all simulation on the CPU.

Qiskit Aer

Qiskit is an open-source framework for working with noisy quantum computers at the level of pulses, circuits, and algorithms. The Qiskit Aer library provides high-performance quantum computing simulators with realistic noise models.

On qBraid, the Qiskit Aer GPU environment comes with the qiskit-aer-gpu package, extending the same functionality of the canonical qiskit-aer package, plus the ability to run the GPU supported simulators: statevector, density matrix, and unitary.

Here is a basic example:

import qiskit
from qiskit_aer import AerSimulator

# Generate 3-qubit GHZ state
circ = qiskit.QuantumCircuit(3)
circ.h(0)
circ.cx(0, 1)
circ.cx(1, 2)
circ.measure_all()

# Construct an ideal simulator
aersim = AerSimulator(method='statevector', device='GPU')

# Perform an ideal simulation
result_ideal = qiskit.execute(circ, aersim).result()
counts_ideal = result_ideal.get_counts(0)
print('Counts(ideal):', counts_ideal)
# Counts(ideal): {'000': 493, '111': 531}

What’s Next

We’re actively expanding GPU capabilities on qBraid. Here’s what to expect:

Dedicated GPU filesystem: GPU sessions currently use a separate high-performance storage backend. Files saved in GPU sessions persist and benefit from faster I/O. Unified filesystem access across all qBraid instances is coming soon.

Expanding environment support: You can create and persist local environments in GPU sessions. Shareable environments and pre-packaged environments from standard qBraid are on the roadmap.

Growing capacity: We’re continuously adding more GPU resources to meet demand. If capacity is temporarily unavailable, check back periodically as capacity becomes available.

Improving startup reliability: If you encounter any NVIDIA setup issues after starting a GPU, run nvidia-smi in a terminal session. If you see an error, simply restart the same instance.