Following the Graviton2's first official announcement back in December, as well as the preview period that has been going on for several months now, Amazon has today publicly launched new m6g instanced based on their new in-house Arm platform.

We had a more in-depth look into the Graviton2 in our analysis article back in March, and the Amazon silicon and Arm's Neoverse-N1 cores certainly delivered on its promises, showcasing competitive performance both in single-threaded as well as multi-threaded scenarios. Most importantly, on AWS, the m6g instances delivered better cost efficiency compared to competing Intel and AMD-based instances. 

The new m6g instances in Amazon's main global regions, including US East (N. Virginia), US East (Ohio), US West (Oregon), Europe (Ireland), Europe (Frankfurt), and Asia Pacific (Tokyo) starting today.

Related Reading:

Source: Amazon Blog

Comments Locked

21 Comments

View All Comments

  • azfacea - Tuesday, May 12, 2020 - link

    going their own way is inevitable. period. full stop. we are going to see more of these from more big players, CDNs and others. sure x86 is not going away. those instances will be available in every cloud, but if the big users like CDNs, netflix, youtube, twitch, instagram, facebook, china, .... increasing move away from x86 in datacenter and make "their own" thats quite a big chunk of x86 server business gone.

    I'd love to see intel, hat in hand, going back to gamers as its primary customers. LUL
  • azfacea - Tuesday, May 12, 2020 - link

    the day pure play foundries succeeded in catching up with intel and really exceeding it, was the day intel went from being a "key partner" to its big customers to being a "business risk", and unnecessary middle man and an uncontrollable bureaucracy with a different focus.

    this does not necessarily happen to gamers but in the data center its just a matter of time.
  • FreckledTrout - Tuesday, May 12, 2020 - link

    Mostly these big cloud providers need multiple sources for everything including CPU's. Its a huge risk to have one vendor of anything. AMD however is really becoming viable which may make Graviton2 mostly pointless. You know I have heard all sorts of crazy rumors but you know I never heard Amazon buying AMD but I'm surprised they didn't when it was cheap a couple years ago. It was probably cheaper than building there own CPU.
  • brucethemoose - Tuesday, May 12, 2020 - link

    Intel/AMD could license out x86 IP like ARM does. AMD's semi custom division is a big step in that direction... and Samsung is already customizing Radeon IP.

    The x86 cross licensing though... I can't even imagine the legal fees involved.
  • amandakspikes - Monday, May 18, 2020 - link

    I’m making more than $75k by just doing very easy and simple online job from home.Last month my friend sister received $94280 from this work by just giving only 2 to 3 hrs a day.Everybody start earning money online. visit for more details… www.iⅭ­a­s­h­68­.Ⅽ­O­Ⅿ
  • MrEcho - Tuesday, May 12, 2020 - link

    Just doing a quick look, they seem to be just a bit cheaper than others of the same CPU/RAM amount, but not by much.
    https://aws.amazon.com/ec2/pricing/on-demand/
  • azfacea - Tuesday, May 12, 2020 - link

    not by much ?

    i remember the graviton1 announcement said compute was 40% cheaper. for compute intensive workloads like video encoding i'd be very surprised it if doesnt beat intel instances by way more than that 40%
  • azfacea - Tuesday, May 12, 2020 - link

    if you are thinking a t3.small has 2 x86 cores (hyptherthreads) just for you. you are reading it very wrong. its a vCPU. if i remember correctly small gets 10% of that core (hyptherthread).

    the x86 cores are just hyperthreads not physical cores, and you only get so much burst credit. the arm cores are full physical cores. i dont know about burst credits on that one.
  • DanNeely - Tuesday, May 12, 2020 - link

    I'm not sure what the non-burst rate on a T instance is; if you're doing enough work for it to matter you probably shouldn't be on a T to begin with.

    An m5.large (1 dedicated x86 core/2threads 8gb of ram) is 9.6 cents/hour. An equivalent m6g.large is 7.7 cents/hour. That's a 20% discount before factoring in any differences in CPU performance.
  • azfacea - Tuesday, May 12, 2020 - link

    you are comparing m5 generation to m6 thats completely flawed. there may be many other features like network, nvme instance store, diff hypervisor, security patches, lots of things that could have changed.

    as for should be using T instance or not its completely diff story. i dont need such education. i was pointing out to him that 2 CPU may not mean just 2 CPUs.

    at the end of the day amazon's 40% cheaper compute claim is not a fabrication on something so easily testable.

Log in

Don't have an account? Sign up now