Comparing Cloud VM Performance Across Architectures Part 2
3 Clouds, 3 CPUs, 3,000 Hours of Benchmarks. A different look at the results.
Let’s pick up from last Wednesday’s post looking at CPU choices and costs based on our recent benchmarks of the three most prevalent options (Intel, AMD and ARM) across the top three cloud providers and around 3000 hours of benchmark runs.
As a reminder these results are based on the COREx benchmark with prices (now slightly out of date) as of 09 Nov 2025 and these results were first presented at Quant Minds 2025. The benchmarks runs included the latest three generations (where available) of CPUs from Intel, AMD and the ARM based options from each hyperscaler (Graviton, Axion and Cobalt).
This time however we’ve selected VMs with at least 2GB of RAM per vCPU (instead of 4), spot prices not on demand, and instead of showing every result there is a small selection. The number in brackets is the position of that CPU/VM in the results if ordered by most cost effective (performance adjusted cost). The charts shows the first occurrence of every type of CPU. (The same CPU may be present in multiple types of VMs and of VMs of different sizes which have different costs and performance characteristic).
Something that struck me about the results (as much as the fact that you really need to be able to use ARM and AMD CPUs) is the pricing strategies that are clearly in play here and how they differ between the three cloud providers.
Lastly, if you’d like to do this kind of analysis on the fly, based on the benchmarks and hardware characteristics of your choice (and integrate that as an API call) we have something you might like coming soon...
