Amazon Web Services (AWS) made an aggressive push into the data center chip space with the unveiling of its purpose-built, Arm-based Graviton2 platform that will power instances within many of its core cloud services. The aggression is set to put pressure on its hyperscale cloud rivals and chip players.

“AWS is dead serious with this new chip, and I don’t say this lightly,” explained Patrick Moorhead, president and principal analyst at Moor Insights & Strategy, in an email to SDxCentral.

AWS CEO Andrew Jassy made the platform unveiling during a lengthy keynote address at this week’s re:Invent show in Las Vegas. He explained that the new chip has a 40% better price performance than the latest generation of Intel-based x86 processors. “That’s unbelievable,” he said in adding gravity to the comparison.

That price performance gain is based on Arm’s 64-bit Neoverse N1 intellectual property (IP) platform that it initially unveiled a year ago. The platform was designed to be optimized for 7nm processor technology and focused on compute speed.

AWS quickly jumped on supporting the platform, working in collaboration with Israeli-based Annapurna Labs, which the company acquired in 2015. Arm officially launched the product earlier this year.

Moorhead said it was “impressive” that AWS was able to roll out this “fully designed and manufactured part” just 10 months after Arm released the N1 IP in February.

“This is not what one expects to see from a silicon design, development, and manufacturing cycle,” Moorhead said. “Literally unheard of. Even assuming AWS got the IP early, which I'm sure they did, this is about as agile a silicon development lifecycle as one can imagine.”

He noted that AWS was likely able to accelerate this timeline because it was not burdened with having to power all data center variations or deal with a lot of legacy support.

From a pure performance perspective, AWS claims the Neoverse architecture can deliver up to 7x the performance of the A1 instances, including twice the floating point performance. It also has additional memory channels and double-sized per-core caches that improve memory access by up to 5x.

“While lots of companies including ourselves have worked with x86 processors for a long time … if we wanted to push the price performance envelope for you it meant we had to do some innovating ourselves,” Jassy said.

That performance will initially be tasked with powering Amazon EC2-based instances, including its EMR big data platform, Elastic Load Balancing, Amazon ElastiCache, and other AWS services. Those instances are built on AWS’ Nitro System, which Moorhead said was AWS’ “secret sauce” for offloading, accelerating, and virtualizing compute, storage, and network services.

“I believe the degree of AWS Graviton2 success will be determined by AWS customer’s degree of translating the performance and cost claims made in its benchmarks above to real customer workloads,” Moorhead said. “AWS has some work to do to bring Graviton2 to all its tools and services.”

Jassy also cited the performance gains as being key to powering more advanced machine learning (ML) applications.

“People talked a lot about training on machine learning, but if you do a lot of machine learning at scale in production … you know that the majority of your costs is actually in the predictions, or the inference,” Jassy said, adding that predictions contribute up to 90 percent of the cost. “This is why we wanted to work on this problem.”

A number of AWS rivals have also targeted ML and artificial intelligence (AI) with updated chip designs, including Alibaba and Google.

AWS’ move could also pressure rival cloud providers to more aggressively embrace alternative chip platforms to what has been typically an Intel play. Arm’s initial Neoverse push was seen as an indirect attempt at scraping away at Intel’s market dominance. Getting AWS on board furthers that push.

Moorhead noted that Arm spent billions of dollars on developing the Neoverse base architecture that will now be leveraged “across numerous manufacturers and further limiting test time and effort. This is unlike other Arm data center plays that developed their own cores. This wasn’t naivety; it was that Arm wasn’t designing and licensing ‘big cores,’” he explained.

The decision making could come down to the real-world price and performance benefits gleaned from rivals. Moorhead specifically pointed to AMD’s second-generation 7nm EPYC processor architecture that it unveiled in August, and Intel’s 10nm Xeon silicon that is expected to launch in the second half of next year.

“Whatever the outcome, we now have even more competition in the data center processor market,” Moorhead said.

Jessica Hardcastle contributed to this story.