AWS caused quite a stir during this year’s annual re:Invent conference when they announced not only were they making their own ARM-based microprocessors called the Graviton 2, but also using those chips to power a new AWS instance type - the EC2 M6g. AWS claims the new instances not only provide several advantages over Intel based processors powering the large majority of the fleet (such as the m5), but that the instance types will also cost less. Truth be told, most customers will find that the cost of a vm is already pretty low, however, savings are savings. Out of curiosity, I decided to dig a little deeper into what that new instance is like during launch week.

Limited Region Support

The first thing customers will notice is that the m6 is only available in certain regions. That’s not new, as Amazon often rolls out services over time. The difference here, is that I’ve only found the m6 in 1 of the 5 US regions as of Dec 4, 2019. Regions are pretty much synonymous with datacenters. Since the new chips are hardware, it makes sense that not every datacenter would have received the new gear. There are some interesting implications here on real-word design. The biggest being a lack of regional failover if you need to spin up an exact EC2 instance based on that machine type.

US East (N. Virginia) - Yes
US East (Ohio) - No
US West (Los Angeles) - No
US West (Northern California) - No
US West (Oregon) - No

CPU and RAM Limits

Quite a bit of information can be found on the EC2 Pricing Page regarding the new EC2 M6g instances ( https://aws.amazon.com/ec2/pricing/on-demand/). The new M6g Instance type has several configurations, however, they simply are not as numerous as the M5 right now. One noticeable difference is that M6g instance types support a maximum of 64 CPUs. That’s a good amount for several workload, but some tasks simply need more power. The M5 can be configured with up to 96 CPUs. Likewise, RAM amounts top out at 256 GB for the M6g while the M5 supports up to 384 GB of main memory.

OS Support

The new chips are capable of running on 2 operating systems - Amazon’s custom Linux, and Red Hat’s Linux. Windows is not an option until Microsoft completes the rumored ARM version of Windows Server.

Pricing

The EC2 M6g Instance type does appear to be less expensive according to the pricing guide (https://aws.amazon.com/ec2/pricing/on-demand/). Just eyeballing the numbers, there appear to be about a 10 - 20% cost difference in some cases.

I’ve included a sample price comparison for the base M6g.Large Instance type against the M5.Large and M5a.Large types.

	m6g.large	m5.large	m5a.large
Number of Processors	2	2	2
RAM (in GiB)	8	8	8
Cost per Hour	$0.077	$0.096	$0.086

Not Available to Everyone

Access to the new M6g Instance type is only available to those who are granted access. Access can be requested on the main page which details the specs of the product offering. The URL is https://aws.amazon.com/ec2/instance-types/m6/?hp=r . I planned to conduct an install, however, I wasn’t granted access before the time of publication. Once I’m granted access, I might do some benchmark tests in the real world.

Final Thoughts

These limitations will not affect everyone in the same way. Most customers will have to wait before considering shifting workloads to the new servers. A single region location and capacity limits will be non-starters for many enterprises. Add to that the need to be approved by Amazon to use the chip, and the hurdles seem too numerous to overcome. Until servers with the new Graviton 2 processor and the M6g Instance types are in wider availability, most customers will be in a holding pattern.

Update 12/08/2019:

I’ve been informed that the Graviton 2 processor supports many more Linux variants than those listed. The OS support includes RHEL, CentOS, SLES, Fedora, Debian, Ubuntu, FreeBSD, and NetBSD.