300-610 · Question #313
An AI/ML team is designing a fraud-detection model for an online banking application. The system must be able to handle a high volume of transaction data, and the models must be trained periodically w
The correct answer is A. GPU-based compute instances connected via high-throughput networking. Real-time inference and periodic retraining of high‑volume ML models demand the parallel processing power of GPUs. Connecting these GPU instances with high‑throughput, low‑latency networking ensures distributed training scales efficiently and that inference calls are processed wi
Question
An AI/ML team is designing a fraud-detection model for an online banking application. The system must be able to handle a high volume of transaction data, and the models must be trained periodically with frequent updates. In addition, inference must happen in real time as transactions occur. The data center must provide low-latency connectivity between compute resources and support scalability as data grows. Which compute solution is most appropriate to meet these requirements?
Options
- AGPU-based compute instances connected via high-throughput networking
- BCPU-based compute instances with local SSD storage
- Ccloud-based compute resources with dynamic elastic scaling
- Ddedicated multi-core CPUs with distributed storage
How the community answered
(54 responses)- A59% (32)
- B6% (3)
- C24% (13)
- D11% (6)
Explanation
Real-time inference and periodic retraining of high‑volume ML models demand the parallel processing power of GPUs. Connecting these GPU instances with high‑throughput, low‑latency networking ensures distributed training scales efficiently and that inference calls are processed with minimal delay. CPU‑only or cloud‑elastic options either lack the necessary parallelism or introduce higher latency, while dedicated multi‑core CPUs don’t offer the same acceleration for deep learning workloads.
Topics
Community Discussion
No community discussion yet for this question.