Cost tracking and saving

With Amazon Braket, you have access to quantum computing resources on demand without upfront commitment. You pay only for what you use. To learn more about pricing, visit our pricing page.

Setting spending limits for Amazon Braket QPUs

Amazon Braket spending limits provide optional per-device cost controls for quantum processing units (QPUs).

How spending limits work: Amazon Braket tracks your cumulative spending and validates every task creation request against your configured limit. If a task's estimated cost exceeds your remaining spending limit, Amazon Braket rejects the task immediately with a validation error. You can optionally configure a time period for your spending limit. By configuring a time period, you can ensure tasks can only be submitted in that specified period. Tasks submitted outside the time period will be rejected.

Opt-in design: Existing workflows will remain unaffected unless you explicitly enable controls. You can remove all restrictions by deleting the spending limit.

List of spending limit actions

While optional, always specify the region parameter as a best practice. Commands executed on a different region than the device's will fail or, in the case of SearchSpendingLimits, return incorrect results.

For more examples on how to use spending limits, see the example notebook.

How task validation works

When the AWS account sends an otherwise valid CreateQuantumTask request, it is subject to the following gating behavior. Note: Remaining budget is the difference between the spending limit and the sum of the queued and current spend. (See next section)

How remaining budget is computed

Remaining budget is the difference between the spending limit and the sum of the current spend and queued spend.

When a task is created for a device with a spending limit, the queued spend is raised by the estimated cost of the task. This event is listed in the first row of the following table. The following table shows what happens to the queued spend and current spend, depending on the progression of the task.

Edge cases

Q: When creating a spending limit, do tasks already in the queue count towards the queued spend?

A: No. Tasks that are already created, queued, or otherwise in-progress do not count towards the queued spend of a newly created spending limit.

Q: Does lowering the spending limit, by updating it, causes early termination of created, queued, or otherwise in-progress quantum task?

A: No.

Q: Does reaching the end time of the spending limit causes early termination of a created, queued, or otherwise in-progress quantum task?

A: No. Created, queued and otherwise in-progress tasks are allowed to complete independent of the spending limit status.

Q: How is the lack of spending limit different from a spending limit of zero dollars?

A: No spending limit allows creating quantum tasks without restrictions. A spending limit with zero dollars blocks all quantum tasks.

Q: Does a spending limit of zero in the past or the future block all quantum task creation?

A: Yes.

Q: When creating a spending limit, will the estimated cost of tasks already in the queue count towards the current spend once said tasks complete?

A: No. Only tasks submitted while a spending limit is active count towards the accumulated spend.

Near real-time cost tracking

The Braket SDK offers you the option to add a near real-time cost tracking to your quantum workloads. Each of our example notebooks includes cost tracking code to provide you with a maximum cost estimate on Braket's quantum processing units (QPUs) and on-demand simulators. Maximum cost estimates will be shown in USD and are not inclusive of any credits or discounts.

Cost tracking for SV1

In order to demonstrate how the cost tracking function can be used, we will be constructing a Bell State circuit and running it on our SV1 simulator. Begin by importing the Braket SDK modules, defining a Bell State and adding the Tracker() function to our circuit:

#import any required modules
from braket.aws import AwsDevice
from braket.circuits import Circuit
from braket.tracking import Tracker

#create our bell circuit
circ = Circuit().h(0).cnot(0,1)
device = AwsDevice("arn:aws:braket:::device/quantum-simulator/amazon/sv1")
with Tracker() as tracker:
    task = device.run(circ, shots=1000).result()

#Your results
print(task.measurement_counts)

Counter({'00': 500, '11': 500})

When you run your Notebook, you can expect the following output for your Bell State simulation. The tracker function will show you the number of shots sent, quantum tasks completed, the execution duration, the billed execution duration, and your maximum cost in USD. Your execution time may vary for each simulation.

import datetime

tracker.quantum_tasks_statistics()
{'arn:aws:braket:::device/quantum-simulator/amazon/sv1':
 {'shots': 1000,
  'tasks': {'COMPLETED': 1},
  'execution_duration': datetime.timedelta(microseconds=4000),
  'billed_execution_duration': datetime.timedelta(seconds=3)}}
  
tracker.simulator_tasks_cost()

Decimal('0.0037500000')

Using the cost tracker to set maximum costs

You can use the cost tracker to set maximum costs on a program. You may have a maximum threshold for how much you want to spend on a given program. In this way, you can use the cost tracker to build out cost control logic in your execution code. The following example takes the same circuit on a Rigetti QPU and limits the cost to 1 USD. The cost to run one iteration of the circuit in our code is 0.30 USD. We have set the logic to repeat the iterations until the total cost exceeds 1 USD; hence, the code snippet will run three times until the next iteration exceeds 1 USD. Generally, a program would continue to iterate until it reaches your desired maximum cost, in this case - three iterations.

device = AwsDevice("arn:aws:braket:us-west-1::device/qpu/rigetti/Ankaa-3")
with Tracker() as tracker:
    while tracker.qpu_tasks_cost() < 1:
        result = device.run(circ, shots=200).result()
print(tracker.quantum_tasks_statistics())
print(tracker.qpu_tasks_cost(), "USD")

{'arn:aws:braket:us-west-1::device/qpu/rigetti/Ankaa-3': {'shots': 600, 'tasks': {'COMPLETED': 3}}}
1.4400000000 USD

Best practices for cost savings

Consider the following best practices for using Amazon Braket. Save time, minimize costs, and avoid common errors.

Verify with simulators

Restrict user access to certain devices

Set billing alarms

Test TN1 quantum tasks with low shot counts

Check all Regions for quantum tasks