Recommended Free Tools
Start with a simulator to debug and explore a circuit; choose a QPU only when you need to test on hardware or study hardware-specific behavior. For ideal simulation, AWS suggests local simulation below 18 qubits, evaluating workload in the 18–24 range, and an on-demand simulator above 24. For noise simulation, its starting points are local below 9 qubits, workload-dependent from 9–12, and DM1 above 12. These are guidelines, not guarantees: circuit operations, host resources, task volume, device compatibility, queues, and current pricing all affect the choice.
Choose by what you need to learn
Amazon Braket offers SDK simulators that run in your development environment, managed on-demand simulators that run in AWS, and quantum processing units (QPUs) that execute on physical hardware. The practical choice depends first on the experiment: debugging an ideal circuit, examining noise, or measuring behavior on a particular device.
- Debug or explore a circuit: start with a simulator. It avoids QPU scheduling and hardware constraints, though its results are not guaranteed to match hardware.
- Model noise: select a density-matrix simulator such as local
braket_dmor managed DM1, within its size limits. - Test actual hardware behavior: choose a QPU whose paradigm and device properties fit the circuit, then account for its queue and cost.
- Use analog quantum evolution: consider QuEra’s Analog Hamiltonian Simulator only if the problem fits its Hamiltonian, register, and control-field formulation; it is not a general-purpose gate-based circuit device.
AWS lists gate-based QPUs from AQT, IonQ, IQM, and Rigetti, as well as QuEra’s analog device. Inventory and availability can change, so use the live Amazon Braket device information when selecting a target.
Use qubit counts as a first filter, not a promise
AWS’s simulator comparison offers these starting points for choosing local versus managed simulation. The ranges are guidance, not a performance guarantee: local speed depends on the computer running the SDK, while cloud task latency and the shape and number of tasks also matter.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minute#1 Best Overall
| Simulation goal | AWS starting point | How to interpret it |
|---|---|---|
| Ideal or standard circuit simulation | Fewer than 18 qubits: local; 18–24: choose based on workload; more than 24: on-demand | Check local memory and compute, circuit operations, and task volume. On-demand task latency can outweigh its capacity advantage for many small jobs. |
| Noisy circuit simulation | Fewer than 9 qubits: local; 9–12: choose based on workload; more than 12: DM1 | Density-matrix simulation grows especially quickly with qubit count. Verify the selected simulator’s documented limit and whether the noise model represents the question you want to study. |
For further detail, see AWS’s simulator comparison and overview of how Braket works. AWS says SV1 runtime rises linearly with gate count and exponentially with qubit count; DM1 generally scales linearly with operations and exponentially with qubits. Qubit count alone therefore cannot predict whether a circuit will finish quickly.
Pick the simulator that matches the task
Local state vector: braket_sv
The SDK’s local state-vector simulator is useful for rapid prototyping and small ideal circuits. AWS documents it for circuits up to 25 qubits, depending on the host machine. It runs in your SDK environment, so available memory and compute—not a cloud allocation—set the practical boundary.
Local density matrix: braket_dm
Use the local density-matrix simulator for small noisy circuits. AWS documents it up to 12 qubits, depending on host hardware. It has the convenience of local execution, but density-matrix resource needs rise steeply as qubits increase.
Rank #2
Managed state vector: SV1
SV1 is AWS’s on-demand state-vector simulator for ideal circuit simulation. AWS’s getting-started documentation describes simulations up to 34 qubits. It is always available and can process multiple circuits in parallel. Shots have a relatively small effect on runtime compared with qubit and operation counts, but this does not remove the need to check task latency and workload fit.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Managed density matrix: DM1
DM1 is the managed option for noisy density-matrix simulation, documented up to 17 qubits. AWS’s simulator-submission guide lists a six-hour maximum runtime, a default of 35 concurrent tasks, and a maximum of 50 concurrent tasks; verify current limits in AWS documentation before planning a large batch. The guide to submitting simulator tasks explains the service limits and submission process.
Managed simulation can accommodate workloads beyond a developer machine’s capacity, but each on-demand task has latency overhead. For a large batch of tiny circuits, that overhead may make local execution the more practical choice. The available information does not establish enough detail to recommend TN1 for this comparison; check current supported-device documentation if tensor-network simulation is under consideration.
Check whether a QPU can run the circuit
QPU choice is not simply a contest in qubit count. Devices differ in paradigm, connectivity, accepted operations, and task constraints. A simulator accepting a circuit does not prove that a QPU will accept it.
- Paradigm: confirm that the device is gate-based or analog in the way your problem requires.
- Supported and native gates: supported gates are accepted by the QPU; native gates can be mapped directly to its control pulses. Other supported gates may need decomposition into native operations.
- Connectivity: compare the circuit’s two-qubit interactions with the device’s connectivity graph. Mapping logical qubits to physical qubits can affect whether and how the circuit runs.
- Shots and task limits: check the device’s current limits against the number of repeated measurements you need.
AWS’s QPU submission example shows compilation to native gates and mapping abstract qubit indices to physical qubits for Rigetti and IonQ devices. That device-specific compilation does not eliminate the need to inspect the target’s gate support, connectivity, and limits. Local simulation also accepts a broader gate set and some OpenQASM features that may not be supported by a QPU or another simulator.
Estimate when the task will run
QPU execution windows, device status, and queue depth vary. QPU and on-demand simulator tasks are queued; QPUs have limited capacity, and execution timing depends on other workloads and device availability. Check the console’s device status, availability windows, and quantum-task and hybrid-job queue depths. The SDK also exposes queue depth and task queue position.
AWS says QPU tasks can be submitted at any time, even if execution windows are limited; a task waits for the device. An offline status can indicate maintenance, an upgrade, or operational recovery. Queue depth helps with planning but does not guarantee a completion time. The SDK’s documented default polling timeout is five days; this is a client-side wait setting, not a promise that the queued task will finish within five days. See AWS’s guide to when a quantum task will run.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Plan cost and shot count before hardware runs
Braket has no upfront commitment, but use is billed. Prices vary by simulator or device and workload; check current Braket pricing and cost-tracking guidance rather than relying on an old example. The SDK Tracker and console estimates can help assess workload-specific costs, but estimates may differ from the bill and may omit other AWS service costs or discounts.
A shot is one repeated execution and measurement. More shots generally improve statistical precision, so base the shot count on the measurement accuracy the experiment needs rather than choosing an arbitrary large number. Before submitting a QPU task, consider expected repetitions alongside its current device price and any related AWS charges.
Best Value
AWS offers optional per-device spending limits for QPU tasks. Its documentation says this control does not cover simulators, managed notebooks, Hybrid Job EC2 costs, or Braket Direct reservations. AWS also recommends billing alerts through AWS Budgets. A simulator run can catch programming and configuration errors before QPU execution without QPU charges, but simulator output need not match hardware results.
A practical selection sequence
- Decide the experiment: use ideal simulation for circuit logic, density-matrix simulation for modeled noise, or a QPU for hardware-specific behavior.
- Apply AWS’s qubit ranges: use the table as an initial local-versus-managed filter, not as a runtime estimate.
- Confirm capacity and workload fit: check the local host or managed simulator limits, circuit operations, number of tasks, and cloud latency.
- Validate QPU compatibility: inspect paradigm, gates, connectivity, shots, task limits, status, and availability for the specific device.
- Check schedule and cost: inspect queue information and current pricing, estimate required shots, and use available tracking or spending controls.
- Run a simulator check first: catch circuit and configuration problems before paying for hardware execution.
For broader terminology and workflow guidance, AWS’s Amazon Braket terms and concepts page defines the service’s main components.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
