One of my clients a couple of years ago was running a legacy LAMP stack on m5a instances, routinely maxing out the CPUs and bleeding serious money every month. Moved them over to c6i instances with the same number of cores, and CPU utilization dropped by 60%.

Lesson: think long and hard before you tie yourself down with multi-year saving plans restricted to an older generation of CPUs.

oh nice, I was actually wondering what was the quickest way to build a meteor.js app (which is mostly a single threaded process).

My 2 cents: I had to scroll down all the way to find out where the "CPU speed" comes from... And I'd still like to know about the "avg" part, so you may want have a more detailed "Method" section (before the results) ;)

This lists r7iz at the top, and AWS says they run at 3.9 GHz. Dunno where this leaves the m5zn and x2iezn family (at 4.5 GHz), or even the z1d (at 4.0 GHz). Frequency isn't everything, but seems strange that they're not to be found in the table.

You're right, just added m5zn. Base frequency is 3.8Ghz and turbo boost 4.5Ghz, but since it is an outdated CPU, they are average at best.

I'll also benchmark the z1d.

z1d is ancient, from 2017. My theory is that they keep it around because of corporates not switching, and wasting $ on slow and more expensive compute?

z1d: Skylake, 4.0 GHz, 2017 X2iezn: Cascade Lake, 4.5 GHz, 2019

Also give us a passmark per dollar or something

Yea was looking for that too. But then the benchmark would have to be multithreaded too I guess.

Does anyone understand why the arm64 instance types are showing roughly half the performance versus the comparable x64 instances in this? Other benchmarks I've seen put them within 20% of each other. Is this just a hyperthreading vs full core thing?