Ampere Computing, the server processor startup headed by former Intel executive Renee James, released new chips based on same architecture used inside almost all the world's smartphones. The company, hoping to take advantage of recent troubles inside Qualcomm and Intel, is entering production with its eyes on cloud computing giants and technology companies with massive data centers.
Ampere, founded last year with the acquisition of Applied Micro’s Arm-based server technology and financial support from the Carlyle Group, announced the Arm-based chips on Tuesday. The new 64-bit chips are based on a 16-nanometer process. The 16-core or 32-core processors run up to 3.3 GHz. The company said that they are being used by Lenovo and other server makers targeting customers as big as Google and Microsoft.
The company, along with competitors such as Qualcomm and Cavium, is trying to weaken Intel’s more than 95 percent market share in server processors. But Arm-based server chips have long lagged behind Intel’s x86 processors in performance. That has derailed almost every previous attempt to take them out of the smartphone world. Whether Ampere’s new server chips can succeed should become clearer over the next year.
Ampere has been mustering troops over the last year for an assault on Intel’s data center business. The company has boosted its headcount from less than 300 to more than 400 people. Ampere also poached about 50 employees from Qualcomm, which has curtailed its ambitions in the server processor market. That decision came amid attempts to stabilize its business during the battle against Broadcom’s hostile takeover launched last year.
The stars may be aligned for the Santa Clara, California-based Ampere. Intel has squandered its process technology lead in recent years. That has opened the door to rivals that outsource production to Taiwan Semiconductor Manufacturing Corporation, including startups like Ampere. Server chips based on Intel’s 10-nanometer node are not likely to be released before 2020, giving data center customers a reason to try something else.
The server market ramped up 43.7 percent to $22.5 billion from last year’s second quarter to this year’s second quarter, according to estimates from International Data Corporation. Servers shipped increased 20.5 percent over the second quarter to 2.9 million units. While volume server revenues jumped 42.7 percent to $18.4 billion, revenues from higher end systems increased 30.4 percent to $1.7 billion, IDC reported.
Using the Arm architecture, Ampere is attempting to undercut Intel with lower performance per watt and performance per dollar. That way it costs potential customers less to operate servers used in storage, internet and cloud applications, vice president of sales and business development Matt Taylor said. “If you think about moving to any new platform, it's work, and customers have to see that there is value to doing that work,” he added.
The power consumption of the new chips based on Applied Micro’s X-Gene 3 technology and targeting customers in the high volume server market is around 125 watts. The 32-core processor supports PCIe 3.0 for high bandwidth I/O communication. It also contains eight DDR4 controllers with slots for 16 DIMMs that can hold a terabyte of memory to address cloud computing workloads. The 32-core product costs $850, while the 16-core model costs $550.
But challenging Intel’s $30-billion server business is an uphill battle because most of the software running inside data centers is intertwined with Intel’s x86 architecture. “Historically, you have these monolithic blobs of software code and porting them to new architectures is challenging,” Taylor said. “Talk to Microsoft. They took Windows and ported it to Arm, but that was not an easy process.” It can also be extremely expensive, he added.
The company—led by C.E.O. Renee James, who managed Intel’s software and services for almost two decades—is trying to simplify the process as much as possible. Ampere has started working with its largest customers directly on software optimization. "This is what Intel used to do,” Taylor said. “It’s a tried and true formula, and we're trying to replicate it. We can't do this with every customer but this is how we think we can accelerate the performance of customer applications.”
The company could also piggyback on Arm's investments in the server space. “Basically, up until now people have been trying to build Arm servers based on mobile cores and the reality is it just would not scale,” Taylor said. But that has changed in recent years as Arm invests more in data centers, hiring Drew Henry last year as VP of its infrastructure business unit. “We don’t have to develop new IP from the ground up,” Taylor told Electronic Design.
Ampere is shifting to TSMC's 7-nanometer process node for its second product line, which is expected to be available in 2019. These chips will have increases in core count, I/O connectivity and memory, Taylor said. They will also support two-socket capabilities, even though the company favors the memory access advantages that a single socket can provide with higher core counts and memory capacity.
“Our roadmap is aggressive and we are going to be coming out with a rapid succession of products,” Taylor, who formerly headed business development for Qualcomm’s server division, said. “We actually have four products on our roadmap for the next few years. They are all funded and in some stage of development or planning.” He added: “What our customers want, and especially what hyperscale customers want, is a multi-generational roadmap.”
