Configuring P-State AMD: m5a.4xlarge



I am playing around AMD's P-States on m5a.4xlarge, to understand how to efficiently tune the ec2 instance to get the price advantage of m5a.4xlarge which is nearly 10% better than Intel. Have you encountered a PoC/experiment where you found a way around it?


asked 2 months ago50 views
1 Answer

Hello, thank you for your post. EC2 instance types that use the Nitro system do not support the ability for an operating system to control the CPU P-state settings. Therefore, the "amd-pstate" driver that is discussed in the documentation you linked cannot be used. Having said that, I can confirm the default CPU P-state settings for these instances types already provide maximum performance. Please feel free to let me know if you have any questions.

answered 2 months ago
  • We ran a fairly simple workload and bought up a burst of "sleep" pods on m5a.4xlarge and m5.4xlarge. To our surprise, we were performing 5.23 worse on m5a.4xlarge. That does not align right with 10% price benefit.

    Further digging allowed me to understand we were spending a lot of cycles on CPU, allowing me to suspect the P-States of the instances. Can you tell me if there are any best practice guidelines to run an m5a.4xlarge?

    Also, do share any docs to give me more info on the CPU Frequency of m5a.4xlarge instances running on Nitro?

You are not logged in. Log in to post an answer.

A good answer clearly answers the question and provides constructive feedback and encourages professional growth in the question asker.

Guidelines for Answering Questions