Quiz: Quantum Cloud Platforms

DataField.Dev

Quiz: Quantum Cloud Platforms

20 questions. Answers with explanations follow.

1. A 4,096-shot Bell-state job occupies the device for:

(a) 6.92 milliseconds
(b) 6.92 seconds
(c) 6.92 minutes
(d) it depends entirely on the queue

2. Local state-vector simulation of the same jobs took:

(a) microseconds
(b) 22–74 ms — the same order of magnitude as hardware execution
(c) several minutes
(d) longer than the queue

3. rz has a duration of zero because:

(a) it is skipped by the transpiler
(b) it is a virtual gate — a phase bookkeeping change, not a pulse
(c) it is only used in simulation
(d) the calibration data is missing

4. At a five-minute queue, a 4,096-shot Bell job has utilization:

(a) 0.5
(b) 2.31 × 10⁻⁵ — 43,340× wall clock over device time
(c) 1.0
(d) undefined

5. ★ Therefore the highest-leverage optimization on a cloud platform is:

(a) reducing circuit depth
(b) batching — submitting many circuits in one job
(c) a better transpiler seed
(d) error mitigation

6. ★ Chapters 28, 29, and 31 optimized circuit depth, layout, and coherence. Relative to §39.3:

(a) they were wasted effort
(b) they improve the 7 milliseconds and do not touch the five minutes — fidelity and throughput are different problems
(c) they eliminate the queue
(d) they only matter on simulators

7. A per-shot price and a per-minute price can be compared:

(a) directly
(b) only with the circuit duration, which is the conversion factor
(c) never
(d) only for identical hardware

8. ★ Chapter 36's VQE run is 18,456,984 shots. In device time that is:

(a) 31.2 seconds
(b) 31.2 minutes
(c) 31.2 hours
(d) 8.6 days

9. ★★★ Priced per minute of QPU time, that run costs:

(a) $50
(b) $7,432
(c) $185,542
(d) $0

10. ★★★ Priced per shot on superconducting hardware, the same run costs:

(a) $50
(b) $7,432 — 149×
(c) $185,542
(d) the same as per-minute

11. ★★★ And on trapped ions:

(a) $50
(b) $7,432
(c) $185,542 — 3,718×
(d) it cannot run

12. ★★ The right conclusion from questions 9–11 is:

(a) trapped-ion providers are overcharging
(b) you are not paying for device time — the bill tracks access, and the unit of access differs by three orders of magnitude
(c) per-minute pricing is always better
(d) the computation is different on each device

13. ★★ The trapped-ion premium buys:

(a) nothing
(b) all-to-all connectivity, longer coherence, better two-qubit fidelity
(c) a shorter queue
(d) free simulation

14. ★★ PriceBook requires an as_of date because:

(a) it is used for sorting
(b) published rates change, and an undated price rots without telling you
(c) providers require it
(d) it identifies the currency

15. ★★ Two-qubit error rates measured across one 133-qubit device spanned:

(a) 1.79e-03 to 1.00 — some links are dead
(b) all within 10%
(c) 0 to 0.001
(d) they are identical by design

16. ★★ Transpiling a 14-qubit circuit with 24 different seeds gave:

(a) identical results
(b) fidelity 0.5755 to 0.7911 — 2.03× the error
(c) failures on most seeds
(d) a 100× spread

17. ★★★ The same experiment on a 4-qubit circuit gave:

(a) a larger spread
(b) exactly zero variation — it fits the coupling map without routing
(c) the same 2.03×
(d) an error

18. ★★★ So a team measuring seed sensitivity on a toy circuit will:

(a) get a good estimate
(b) conclude the seed does not matter, and be wrong for every circuit they care about
(c) find the same answer more cheaply
(d) overestimate the variance

19. ★★ ExecutionRecord.is_reproducible is False for a record with a pinned environment and a fixed seed because:

(a) the seed is not enough entropy
(b) four fields that determine the result come from the provider at execution time and are not in your code
(c) the version string is wrong
(d) hardware is never reproducible

20. ★★★ 120 VQE iterations at a five-minute queue, submitted as 120 jobs:

(a) 10 hours waiting, under a second computing
(b) 5 minutes total
(c) 120 hours
(d) the same as one batched job

Answers

1. (a) 6.92 ms. From real calibration data: 1.69 µs per shot × 4,096.

2. (b) 22–74 ms. Both are tens of milliseconds. Everything that makes hardware feel slow happens around the execution, not during it.

3. (b) a virtual gate. Chapter 31 measured why: a $Z$ rotation is absorbed into the phase of subsequent pulses rather than being played.

4. (b) 2.31 × 10⁻⁵. The processor spends 99.998% of your job's lifetime serving somebody else.

5. (b) batching, worth ~99× for 100 circuits — more than any transpiler setting. And it is not a quantum technique. The first question about a slow quantum workflow is "how many jobs?"

6. (b). They optimized fidelity; this chapter is about throughput. Both are real, and a team that conflates them optimizes the wrong one.

7. (b). A per-shot model charges the same for a 1.69 µs Bell circuit and a 10.55 µs QFT; a per-minute model charges 6× more for the second. Neither is wrong — they price different things.

8. (a) 31.2 seconds. Eighteen million shots sounds enormous and is half a minute of hardware.

9. (a) $50.

10. (b) $7,432 — 149×, including the per-task fee across 3,240 tasks.

11. (c) $185,542 — 3,718×.

12. (b). Per device-second: $2, $238, $5,948, for identical computation. This is the practical form of a pattern the book keeps finding — Chapter 36's error was dominated by an approximation nobody reported, Chapter 37's comparison turned on what an algorithm produced rather than how it scored. The quantity that determines your invoice is not the quantity anyone optimizes.

13. (b). So 28× per shot is not a markup for nothing. But it means a shot count chosen on superconducting hardware is a budgeting decision, not a physics decision, when moved — which is Case Study 39.1's entire subject.

14. (b). A hard-coded number silently rots; a dated one tells you when to re-check it.

15. (a). Some links have error 1.00 — they are dead — and T1 spans 15.2 to 483.0 µs. Which physical qubits you get is a first-order determinant of your result, and the platform assigns them at execution time.

16. (b) 2.03× the error, driven by a 49-to-112 spread in two-qubit gate count. That is routing overhead: the transpiler inserts swaps to satisfy connectivity, and how many depends on where it started.

17. (b) exactly zero. A 4-qubit chain fits the coupling map without any routing, so the transpiler finds the same layout every time.

18. (b). And note the shape of the error: the 4-qubit measurement is a true measurement of a false general claim. This is the book's seventh instance of a measurement being too small — and the first where "too small" means the wrong system rather than too few samples.

19. (b). Job ID, execution timestamp, calibration snapshot, and physical qubit assignment. A pinned requirements.txt and a fixed seed pin the client side of a computation whose variance lives on the server side. If you did not capture them at execution, they are gone.

20. (a) 10 hours waiting, under a second computing. In a session the queue is paid once — five minutes. That is what sessions are for, and it is the difference between a workflow that runs overnight and one that runs over lunch.