At AWS Re:Invent the company is making a number of key announcements about new instances as expected. Aside from highlighting the new bare metal i3.metal instance public preview, there is another instance type available today that will have a far bigger impact: the AWS EC2 M5 instance type.
Amazon AWS EC2 M5 Instance Details
Here are the details of the new Amazon AWS EC2 M5 instance types:
||Network Bandwidth||EBS-Optimized Bandwidth||$/hr|
|m5.large||2||8 GiB||Up to 10 Gbps||Up to 2120 Mbps||0.096|
|m5.xlarge||4||16 GiB||Up to 10 Gbps||Up to 2120 Mbps||0.192|
|m5.2xlarge||8||32 GiB||Up to 10 Gbps||Up to 2120 Mbps||0.384|
|m5.4xlarge||16||64 GiB||Up to 10 Gbps||2120 Mbps||0.768|
|m5.12xlarge||48||192 GiB||10 Gbps||5000 Mbps||2.304|
|m5.24xlarge||96||384 GiB||25 Gbps||10000 Mbps||4.608|
Like with the recently launched AWS C5 instances for compute, the new M5 instances rely upon a custom Intel Xeon Platinum 8175M CPU. That means that these new instances can have up to 14% better performance per core for general purpose workloads and up to 2x the FLOPS performance with AVX-512 enabled code.
All is not perfect with the new instances. From the launch announcement:
One quick note: the current NVMe driver is not optimized for high-performance sequential workloads and we don’t recommend the use of M5 instances in conjunction with sc1 or st1 volumes. We are aware of this issue and have been working to optimize the driver for this important use case.
Hopefully this gets fixed soon.
For those building there own infrastructure, there are some implications. First, the network and storage bandwidth offered by the new EC2 M5 instances are low, and what many of our readers are running in their home lab environments. Second, Amazon is all-in with high spec CPUs, high-density memory, high-spec storage and >10GbE networking. As we move into 2018, building your own infrastructure is going to mean you will want more NVMe, more 40GbE+ networking and higher memory capacities. The AMD EPYC becomes extremely compelling here because one can use 1TB or more of RAM without having to move to the M parts. AWS custom CPUs are well known and expected, but it is also known that AWS is a company that gets heavily discounted components from Intel and others. That is something to keep in mind when planning 2018 and later infrastructure.