Hey everyone!
We're in the process of migrating some on-prem VMs to AWS and are considering making the switch from Intel instances to AMD types. Historically, we've only used Intel instances, but I'm curious if anyone has successfully transitioned to AMD and what the performance differences have been.
I've heard that A instances might be cheaper than I instances, but my research shows mixed results:
- c7i.xlarge costs .3625 while c7a.xlarge is .3893
- m7i.xlarge costs .3856 while m7a.xlarge is .4158
Older generation instances do show AMD as slightly cheaper, like m6i.xlarge at .376 and m6a.xlarge at .3568.
Since we're under pressure to save on costs, can someone clarify: Are AMD instances only a better deal on older models? Are the newer AMDs faster than their Intel counterparts, allowing for downsizing without a performance hit?
I've had experiences where newer Intel instances (like m7i-flex) were cheaper than their predecessors, which adds to my confusion!
Any insights or experiences you could share would be super helpful. Thank you!
4 Answers
Yeah, measurement is key. The AMD 5a and 6a series tend to have higher memory latency compared to Intel in the same generation. If you're really looking to cut costs, consider using the Graviton 7g or 8g instances instead.
If your workloads allow for it and you're not tied to Windows, switching to arm64 Graviton instances could lead to significant savings. They're typically cheaper than AMD or Intel instances too.
The best way to determine performance is to run your own tests. Every workload can behave differently based on its setup. Just try migrating some workloads and see how they perform on AMD vs Intel.
When moving VMs to AWS, it might not save you a ton in operational costs. It might be better to rethink if the workloads should be refactored into different technologies like containers or lambdas. But for what it's worth, AMD instances have generally performed well in most scenarios I've seen.
That's a good point. As we shift away from data centers, I'm mostly focused on ensuring we're not overprovisioning our instances and wasting money.