We had previously written about Amazon’s announcement in December last year to provide a quantum cloud service.  They have been running the service in a beta test mode to selected customers for the past several months and now have announced General Availability for any AWS customer. The service is currently offering access to the D-Wave 2000Q quantum annealing processor, the Rigetti Aspen-8 superconducting gate level processor, and the IonQ ion trap processor. In addition, the service also includes a high power state vector simulator (SV1) that can simulate circuits up to 34 qubits on AWS classical cloud computers.  Along with the hardware and simulators, AWS is also providing its own Amazon Braket Python SDK that is an open source library to design and build quantum circuits and submit them to Braket quantum processing units (QPUs) as tasks and monitor their execution.  The SDK also includes a lower performance quantum gate simulator that can handle up to 25 qubit simulations and runs on the user’s notebook.

AWS divides up their computing resources into regions and currently has each quantum processing unit (QPU) attached to a specific region.  These include U.S. West (Oregon), U.S. West (N. California), and U.S. East (N. Virginia). The actual quantum processors are located at the providers facilities and the jobs are sent to the providers over high speed lines in anonymized and encrypted form. The results sent back to the AWS data center over the same lines and the AWS infrastructure handles any necessary job queueing, jobs status notification, and storage of the results in the AWS S3 storage service.

Of particular interest to us is the pricing table that AWS provided as follows:

Hardware ProviderQPU familyPer-task pricePer-shot price
D-Wave2000Q$0.30$0.00019
IonQIonQ device$0.30$0.01000
RigettiAspen-8$0.30$0.00035
AWSSV1 Simulator$4.50 per hour

A task is one job with the same circuit configuration and a shot is a repetition of the calculation.  The pricing is independent of the number of qubits used as well as the number of gates in the circuit. Typical tasks consists of somewhere between 1,000 and 10,000 shots to provide a good statistical base for the results of the calculation. So the cost of executing a single task will have a minimum of $0.30 and could cost as much as $10 or more.  It is notable that the per-shot price of the IonQ processor is much higher than the Rigetti processor. This is because the inherent gate delays in an ion trap processor are much slower than a superconducting processor.  But the potential qubit quality factors of an ion trap design can be much higher and it could potentially provide better results.

AWS has provided a great deal of documentation about this service and here are links to several of the documents.

August 14, 2020