Amazon Talks Up Big Performance Gains For Their 7nm Graviton2 CPUs

[phoronix]

Phoronix: Amazon Talks Up Big Performance Gains For Their 7nm Graviton2 CPUs

We weren't too enthusiastic about the performance of Amazon's initial Graviton ARM-based CPU cores offered via their Elastic Compute Cloud, but their next-gen Gravin2 CPUs that are "coming soon" should be much more capable for good ARM Linux performance...

http://www.phoronix.com/scan.php?page=news_item&px=Amazon-Graviton2-Coming

[uid313]

Sounds cool, I wish these were available for consumers to buy instead of just rent in the cloud.

[bridgman]

but their next-gen Gravin2 CPUs that are "coming soon"

Minor typo...

[CommunityMember]

Unfortunately, still no lightsail graviton instances available for people to easily get their feet wet with minimal friction (sure, anyone with a bit of experience with AWS can spin up a full new instance with storage and networking in minutes, but lightsail makes it much easier, and often cheaper, too, for small sandboxes).

[milkylainen]

So I guess the big question will be raw computing power vs power consumption vs cost as usual?

Also. Why would Amazon do this? I'm not sure I understand a why.
They have shitloads of cash to spend on fun projects?
Lessen reliance on Intel and AMD?
Implementation introspection for security purposes?

[_Alex_]

Why would Amazon do this? I'm not sure I understand a why.

To wake up x86 manufacturers that there are better alternatives in terms of cost and perf/watt. If they start losing clients maybe they'll start making better CPUs.

[coder]

So I guess the big question will be raw computing power vs power consumption vs cost as usual?

Well, their pricing it at "40% better price/performance than x86 instances". How that translates into the underlying costs we'll probably never know. However, the intrinsic perf/W advantages of ARM vs. x86 have been long- and well- established.

Also. Why would Amazon do this? I'm not sure I understand a why.

Between purchase cost and power efficiency, perhaps it's actually cheaper for them to make their own chips? They're using off-the-shelf ARM IP for the cores & probably most of the rest of it.

The chips were designed by Annapurna Labs, which Amazon bought in January 2015. Back then, there weren't really compelling ARM-based server options readily-available. If they were eyeing such an acquisition today, I wonder if they wouldn't instead just source off-the-shelf chips like Cavium's Thunder X2 or Ampere's eMAG.

[coder]

To wake up x86 manufacturers that there are better alternatives in terms of cost and perf/watt.

$350M - $370M is a very expensive wakeup call. And that was just Annapurna Labs' supposed purchase price. Add in the costs to run the group for nearly 5 years and produce several generations of silicon and you're talking about some significant fraction (or more than) $1B.

Companies don't spend that kind of money "to send a message". They do it because there's a business case -- a return-on-investment -- some way they can not only recoup the cost, but further enhance their profitability. And not by simply betting on the reactions of their suppliers, but in a way they can control more directly.

[_Alex_]

$350M - $370M is a very expensive wakeup call. And that was just Annapurna Labs' supposed purchase price. Add in the costs to run the group for nearly 5 years and produce several generations of silicon and you're talking about some significant fraction (or more than) $1B.

Companies don't spend that kind of money "to send a message". They do it because there's a business case -- a return-on-investment -- some way they can not only recoup the cost, but further enhance their profitability. And not by simply betting on the reactions of their suppliers, but in a way they can control more directly.

Well, AWS has 26 billion annual revenue with 7 billion being profit, so they can just put it in the expenses to decrease hardware costs, decrease operating costs and increase future profitability.

[coder]

Well, AWS has 26 billion annual revenue with 7 billion being profit, so they can just put it in the expenses to decrease hardware costs, decrease operating costs and increase future profitability.

I think it's a given that there was a business case for it. I'm just saying that the business case must've been more than simply "to put pressure on their suppliers". These chips would have to deliver cost savings or competitive advantage, in their own right.