Amazon overhauled the low end on July 2014, basically replacing t1.micro with three T2 types: t2.micro, t2.small and t2.medium. These new instance types also introduce a bit of an overlap with other lower-cost types. The difference with t2 and other current instance types is that t2 is designed (and priced) to run mostly idle, with an occasional burst of processing power needed every once in a while. This is just the thing you'd need for a web server, and works fine for many other roles too.
Here's a summary of the low end of EC2 instances:
| Cost per hour | Cost per month | |
|---|---|---|
| t1.micro | $0.02 | ~$14.60 |
| t2.micro | $0.013 | ~$9.49 |
| t2.small | $0.026 | ~$18.98 |
| t2.medium | $0.052 | ~$37.96 |
| m3.medium | $0.07 | ~$51.11 |
All costs in this article are from December 2014 at the US East region.
Note: These costs don't include EBS (a virtual HD) or Data Transfer costs, which are a mandatory addition in this case. An alternative to EBS is the Instance Storage, but that's not applicable to t1 or t2 instances and is outside the scope of this article. The smallest possible (8 GB) EBS volumes cost $0.4 to $0.8 / month to store depending on the volume type. There's a cheaper version, Magnetic volume, which is now practically superseded by General Purpose SSD volume. Since the difference in cost is so small when compared to the actual instance cost, paying for the SSD option is a better choice. Data transfer is the other extra cost that you should be aware of, but usually isn't a problem since the first 1 GB / month of outbound traffic is free and it starts at $0.09 / GB after that.
t1.micro vs t2.micro
Both micro instances work with the same principle, but the old t1.micro worked in a more undocumented and obscure way when compared to the new t2 instances' CPU Credit model. See: T1 Micro Instances in AWS Documentation and T2 Instances in AWS Documentation.
Here's a highlight of the main differences:
| t1.micro | t2.micro | |
|---|---|---|
| Virtualization | PV | HVM |
| 32-bit | Yes | No |
| Memory | 0.615 GB | 1 GB |
| Cost per hour | $0.02 | $0.013 |
What these pretty much tell is that t2.micro is a newer generation instance, and t1.micro is still around for legacy deployments.
t2.medium vs m3.medium
Both of these instances have roughly the same amount of memory (4 GB for t2.medium, 3.75 GB for m3.medium). This comparison boils down to t2 instances' dynamic CPU allotment vs other instances' fixed CPU power. If you look at the CPU Credit comparison table at AWS documentation, you'll see that t2.medium is very different from t2.micro in the way its CPU Credits are used and accumulated.
You'll probably know what your server is going to run and how resource-intensive it will be, but here's a general suggestion: go with t2.medium and set up a CloudWatch alarm to monitor its CPU Credits.
Conclusion
You might've already come up with a conclusion based on the cost comparison table at the start of this article. Going with the lowest-cost t2.micro is actually a good choice. If you're launching a new instance to run a few WordPress sites or something similar, go with t2.micro and scale up to t2.small or t2.medium if needed. If you still have existing t1.micro instances, migrating to the cheaper and better t2.micro only makes sense.