One of the buzziest chip startups in recent memory is ready to start shipping its Arm-based data center processor, in hopes of carving out a foothold in a market dominated by Intel.
Ampere plans to announce later on Tuesday that it is shipping two varieties of its eMAG server processor, which could finally give backers of the Arm chip architecture a few wins in the data center market. Lenovo plans to release a server based on the processor, and cloud computing vendors will be able to obtain the chips in volume production to help determine if there’s room for another company inside the data centers of the world.
Led by former Intel president Renee James, Ampere is attempting to do what several companies have tried in the past: advance an alternative chip architecture within the data centers of the world, which are run almost exclusively on x86 processors designed and built by Intel. Chips based on cores designed by Arm are found in nearly every mobile phone on the planet and quite a few internet-of-things devices, but despite attractive power consumption features, they have yet to make an impact among buyers of large quantities of servers.
Matt Taylor, senior vice president of sales and business development of Ampere, thinks the company is ready to step up and challenge Intel in this market.
“Our customer focus is very much the hyperscale customers, and you know who those are,” Taylor said, likely referring to what Intel executives have called the Super Seven server chip buyers: Amazon Web Services, Microsoft, Google, Facebook, Alibaba, Tencent, and Baidu. “We have deep engagements with all of them, and are working diligently with them on hardware and software,” he said.
The startup, backed by The Carlyle Group and built on the server assets of Applied Micro Circuits, is now shipping two server chips at different power envelopes that were built on TSMC’s 16-nanometer manufacturing process, Taylor said. A 32-core version will consume 125 watts of power running flat-out at 3.3GHz, which compares favorably to the range of Xeon server processors offered by Intel. Actual performance comparisons will have to wait until the chips have been tested by third parties.
These eMAG chips are really a proof-of-concept step for Ampere, given that they only support single-socket servers at the moment. Dual-socket chips built on TSMC’s 7-nanometer manufacturing process — which is currently considered the state of the art in chip manufacturing as Intel struggles with production problems — will probably arrive next year, although Ampere isn’t committing to a ship date for those second-generation chips.
Ampere has hired aggressively over the last several months, Taylor said, snapping up a team of engineers that were working on Qualcomm’s ill-fated server chip efforts and assembling a roster of chip design veterans from companies across the industry. “They’re bringing in a lot of great ideas on design philosophy, and how we can be more agile and develop a rapid cadence of products that are leadership products in the market.”
Taylor thinks that rapid cadence of development is a key advantage for Ampere: it went from launch to shipping product in eight months, which is pretty quick by chip development standards. Chip design cycles have typically taken at least a year, and often two years, which is an eternity in a market that is changing as quickly as modern enterprise computing.
But the real challenge for Ampere will be getting the server market, which has been writing software around the x86 architecture since the Clinton administration, to embrace a new architecture. Arm’s appeal has always been centered around its performance-per-watt characteristics, which make it ideal for small devices like mobile phones and which has also appealed to companies running massive data centers that are spending a lot of money on electricity to power and cool their servers.
Several key enterprise software vendors, including Oracle, Microsoft, and VMware, have announced support for the Arm architecture. But there are countless enterprise applications that were designed around Intel’s chips for decades that will be difficult to move to an alternative architecture, no matter how compelling the power characteristics.