The thing is though for many workloads (I'd even say most) applications get I/O constrained way quicker than CPU, so scaling CPU will not give you the boost that you expect on AWS and the other big cloud providers.
Keep in mind that PCIe4 NVMes can do 5-7GB/sec (gigabytes - not gigabits) read and write, with up to 500K IOPS.
Compare that to RDS on AWS which charges a comical $0.10 per IOPS-month, so just for IO it would be $50k/month (and for many DBs IOPS are more important than CPU) to meet the performance of one NVMe SSD, which can be bought retail for probably $200.
I don't think I've ever seen such ridiculous pricing. I do understand AWS can be more expensive, but it is just outrageous margin for IO performance (and external bandwidth, but that's a whole other story!).
Keep in mind that PCIe4 NVMes can do 5-7GB/sec (gigabytes - not gigabits) read and write, with up to 500K IOPS.
Compare that to RDS on AWS which charges a comical $0.10 per IOPS-month, so just for IO it would be $50k/month (and for many DBs IOPS are more important than CPU) to meet the performance of one NVMe SSD, which can be bought retail for probably $200.
I don't think I've ever seen such ridiculous pricing. I do understand AWS can be more expensive, but it is just outrageous margin for IO performance (and external bandwidth, but that's a whole other story!).