Amazon AWS has two new instance types with its Graviton2 Arm-based processors. The EC2 M6g general compute instances were the first to hit GA, now Amazon is starting to roll-out more instance types to build an ecosystem. The C6g series is Amazon’s compute-focused instance type while the R6g are higher-memory instance types. One can see why they would be released together from a resource partitioning standpoint.
AWS EC2 C6g Instances
Here is a quick table of the Amazon AWS EC2 C6g instances as they hit GA today:
As one can see, we have medium instance types with Graviton2. This is likely to help foster the adoption of the new instance types with even lower-cost instances. We can compare these to the AWS EC2 C5a instances that just hit GA and see that AWS is providing a discount for the new Arm-based servers. It is also offering more EBS bandwidth and network bandwidth. That may also be due to being a C6 instance type instead of a C5 instance type.
When we look at the EC2 C5 instances, we can see a similar trend.
There is a fairly large delta now between the offerings with the Arm instances being the cheapest, Intel is the most expensive, and AMD is in the middle.
There are also C6gd instances with locally attached NVMe storage that will be available soon.
AWS EC2 R6g Instances
On the AWS EC2 R6G side we see something similar. Here is that instance table:
R instances are generally for higher-memory workloads. For example, a Redis or Memcached server where one does not see an enormous CPU burden but performance and capabilities are much more closely tied to the amount of available memory.
Like the C6g instances, local NVMe-backed storage R6gd instances are coming soon.
Final Words
It makes sense that AWS is rolling out these new instance types as a pair. The C6g instances have a higher CPU to memory mix while the R6g has a higher memory to CPU mix. That helps fit and schedule VMs on the physical hardware versus only having general purpose and memory-optimized instance types available.
Overall, this is great for Arm server momentum. It is also a big deal for those looking to do Arm-on-Arm development. We tested and published our AoA Analysis where a Marvell ThunderX2 Equals 190 Raspberry Pi 4. While the ThunderX2 was launched over two years ago (Patrick our Editor-in-Chief spoke at that launch), the Graviton2 is a more modern Neoverse processor. In the end, we now have three processor options on AWS which is a stark contrast to other hyper-scalers like Facebook that have remained heavily invested in the Intel ecosystem.
wow, 64 threads/128gb ram for $1500/mo. double-triple that for the traffic cost. thats expensive for what’s x86 2U colo server can do.
@alt try renting that server by the minute at your local colo. The advantage of cloud providers is not the lowest cost for resources but the dynamic assignment of resources when needed.
Pricing in the article is also the on demand pricing, there are significant discounts at all the cloud providers when you commit to a certain level of usage.