Further thoughts on AWS Graviton2


Turning a Twitter thread into a post.

I wrote about the performance of AWS’s Graviton2 Arm based systems on InfoQ

The last 40 years have been a Red Queen race against Moore’s law, and Intel wasn’t a passenger, they were making it happen. I used to like Pat Gelsinger’s standard reply to ‘when will VMware run on ARM?’

It boiled down to ‘we’re sticking with x86, not just because I used to design that stuff at Intel, but because even if there was a zero watt ARM CPU it wouldn’t sufficiently move the needle on overall system performance/watt’. And then VMware started doing ARM in 2018 ;)

Graviton 2 seems to be the breakthrough system, and it didn’t happen overnight – they’ve been working on it since (before?) the Annapurna acquisition. AWS have also been very brisk in bringing the latest ARM design to market (10 months from IP availability to early release).

Even then, I’m not sure that ARM is winning so much as x86 is losing (on ability to keep throwing $ at Moore’s law’s evil twin, which is that foundry costs rise exponentially). ARM (and RISC-V) have the advantage of carrying less legacy baggage with them in the core design.

I know we’ve heard this song before for things like HP’s Moonshot, but the difference this time seems to be core for core the performance (not just performance/watt) is better. So people aren’t being asked to smear their workload across lots of tiny low powered cores.

So now it’s just recompile (if necessary) and go…

Update 20 Mar 2020

Honeycomb have written about their experiences testing Graviton instances.

No Responses Yet to “Further thoughts on AWS Graviton2”

  1. Leave a Comment

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.

%d bloggers like this: