[singhrac]

As a separate data point, I briefly switched one of our servers from an r6a.4xlarge (AMD Epyc) to a r6i.4xlarge (Intel Xeon) and saw a 30% speedup in our number-heavy compute task. I would love to find out why (MKL or AVX512? Do I need to recompile numpy?), but for the time being it pays to stay on Xeon.

We eventually switched to m-instances since that fits our compute/memory usage better when we’re at limits.

[re-thc]

If it's AVX512, you'll be excited to know that m7a, currently in preview has great support for this. AMD Epyc 4th gen i.e. genoa is a lot faster.

[harpratap]

Yes, I point this out in the article too. Which CPU will perform better is heavily dependent on your workloads, so I refrain from relying 100% on synthetic benchmarks and directly ran canaries in production instead. It's definitely possible Ice Lake is superior for your workload than Milan