Compute size is measured in CUs. 1CU is 1vCPU and 4GiB of RAM. Database compute and it's read replicas autoscale from 0.25 to 10 CU based on load and down to zero when inactive.
> Compute size is measured in CUs. 1CU is 1vCPU and 4GiB of RAM.
Sure, but that precisely doesn't answer the question.
What is 1vCPU in this 1CU? If I benchmark this CPU for example, what do the numbers look like?
I saw discussion comparing the cost of Neon vs Supabase based on CPU but felt it lacking without any indicator on what is a CPU. 1 CPU could be up to 2x the other CPU.
As in the Fargate / Lambda vCPU? There's x86 (old) and arm64 (graviton 2).
As in an EC2 vCPU which as indicated could range from t2 all the way to m7i / m7a. Even just comparing AWS' own graviton from 2 to 4 (latest), you get about 2x performance improvements.
If we compare to RDS / Aurora, it lets you pick the CPU, so definitely makes a difference.
Does this specific detail matter if the customer can just scale up (big number mean faster ug!) and it meets their requirements? Also I'm sure this has changed and will continue to change over time, so it's not exactly easy for the CEO to come on here and tie the infra engineers hands by publishing the exact details of whatever they currently have deployed at the moment just for some hardass to bust their balls about it later. If you really care that much ($$$) I wouldn't be surprised if they'll let you use whatever fancy pants core you have in mind.
> Does this specific detail matter if the customer can just scale up
Yes, as it drifts further and further the scaling can be meaningless. This is exactly what's lost with the newer generation and cloud.
Also with the databases mentioned it can't be horizontally scaled. Writes are single server. So there's a limit.
> so it's not exactly easy for the CEO to come on here and tie the infra engineers hands by publishing the exact details of whatever they currently have deployed at the moment
Nothing to do with infra. This is about the product itself. What are you selling? That's the question. The product could be very expensive or not based on that. It also might not meet any organization needs e.g. since they have to meet a performance criteria and that's not just about more CPUs.
> If you really care that much ($$$) I wouldn't be surprised if they'll let you use whatever fancy pants core you have in mind.
And likely the competition can too. Point being? Does that matter? Imagine going to a store and buying a "shirt". You might be allergic to cotton. Now do you just buy any shirt hoping it works?
Within Fargate you even get different capabilities too, we were having mixed performance between instances and eventually stuck a log statement in to `cat /proc/cpuinfo`. We got different generations of Intel CPU between containers, from Broadwell to Haswell. You had no control over this when provisioning it.
And autoscaling works like this: https://neon.tech/docs/introduction/autoscaling