MLS-C01 · Question #331
A company builds computer-vision models that use deep learning for the autonomous vehicle industry. A machine learning (ML) specialist uses an Amazon EC2 instance that has a CPU:GPU ratio of 12:1 to…
The correct answer is C. Use memory-optimized EC2 Spot Instances for the training jobs. To reduce deep learning training costs when the GPU is underutilized without increasing training duration, leveraging cost-effective Spot Instances is the most direct solution.
Question
A company builds computer-vision models that use deep learning for the autonomous vehicle industry. A machine learning (ML) specialist uses an Amazon EC2 instance that has a CPU:GPU ratio of 12:1 to train the models. The ML specialist examines the instance metric logs and notices that the GPU is idle half of the time. The ML specialist must reduce training costs without increasing the duration of the training jobs. Which solution will meet these requirements?
Options
- ASwitch to an instance type that has only CPUs.
- BUse a heterogeneous cluster that has two different instances groups.
- CUse memory-optimized EC2 Spot Instances for the training jobs.
- DSwitch to an instance type that has a CPU:GPU ratio of 6:1.
How the community answered
(35 responses)- A6% (2)
- B3% (1)
- C83% (29)
- D9% (3)
Why each option
To reduce deep learning training costs when the GPU is underutilized without increasing training duration, leveraging cost-effective Spot Instances is the most direct solution.
Switching to CPU-only instances for deep learning models, which heavily rely on GPUs, would significantly increase the training duration, violating the requirement.
Using a heterogeneous cluster adds complexity and does not directly address the issue of GPU underutilization or guarantee cost reduction for a single training job without increasing duration.
Using memory-optimized EC2 Spot Instances for training jobs provides significant cost savings (up to 90%) compared to On-Demand instances without altering the computational performance or duration of the training job itself, provided the job can handle potential interruptions.
Switching to an instance type with a CPU:GPU ratio of 6:1 (more GPU relative to CPU) when the GPU is already idle half the time would likely increase costs due to higher GPU spend without necessarily improving utilization or reducing overall cost for the task.
Concept tested: Optimizing EC2 instance costs for ML training
Source: https://aws.amazon.com/ec2/spot/
Topics
Community Discussion
No community discussion yet for this question.