In a little over a year since former Intel president Renee James founded Ampere Computing, the semiconductor startup has released its first Arm-based data center chip and is on track to roll out its 7nm chip next year. It’s also partnering with Packet, Cloudflare, and other companies on 5G and edge computing proof of concepts (PoCs).
“We’re involved in a bunch of PoCs in this space,” said Matt Taylor, Ampere’s SVP of world-wide sales and business development. “We want to learn the application space, what the workloads are, and build a product that is more targeted for what those customers and deployments need. Cloudflare has done a lot of public work around Arm, and they are arguably one of the largest edge companies today. That’s another good example of a customer moving rapidly toward Arm servers at the edge.”
It’s been a pretty rapid ascent for the Silicon Valley startup. Ampere’s first Arm-based processor, announced in September, went from IP to product in nine months. “It’s pretty doggone impressive for the team to be formed and to get a product out into mass production in under a year,” Taylor said. “In the semiconductor world, that is light speed.”
The product is a 32-core processor with eight memory controllers and 42 lanes of PCIe 3.0 for high bandwidth input-output (I/O). This large memory footprint is important for in-memory databases, storage, and other applications that require high memory and core count.
Taylor says Ampere’s chip is comparable to Intel’s Xeon Gold processor family and AMD’s mid-level chip (it has three families of processors).
Lenovo is already using this chip in its servers. Taylor can’t publicly name other customers. But Ampere’s going after the top cloud providers. “We’re very focused on the largest hyperscalers, the super 8 or super 9,” he said. “We basically have engagements with every one of them. We’re at different phases with each one, but the vast majority are sampling, testing, doing PoCs on our platform today.”
The startup’s next-generation chip, due in 2019, will target hyperscale cloud and edge computing. These chips will use Taiwan Semiconductor Manufacturing Co.’s (TSMC’s) 7nm technology and will have with multi-socket and single socket options.
“We’re basically enhancing every aspect of the product: more CPU cores, more memory, more I/O, faster speeds,” Taylor said. “Everything will go up and to the right. And then beyond that, we have two other products we are developing on the Arm instruction set.” He won’t say much about these other than they will be “as much as possible on the leading edge, assuming it is not the bleeding edge.”
Ampere acquired IP and assets from Macom, formerly AppliedMicro, one of the original Arm-based server vendors. Macom put the server assets up for sale, and they came across James’ desk at the Carlyle Group. “The private investment firm said we don’t do semiconductor investments, but she wouldn’t take no for an answer,” Taylor said.
The firm finally agreed, but on the condition that James would run the new company. So James spearheaded the investment, bought the assets, and formed Ampere. She also convinced other senior Intel engineers to come out of retirement and work for the startup. Its leadership team now includes former Intel, AMD, Cavium, and Qualcomm executives.
One Year In
A year later Ampere has about 400 employees. It’s headquartered in Santa Clara, with offices in Portland, Oregon, and Raleigh, North Carolina. It also has development centers in India and Vietnam — both former AppliedMicro facilities.
The company won’t say how much money it has raised. But Taylor says it’s enough to fund “multiple generations” of products. “We have four on the roadmap today and we’re well-funded enough to execute those without having to raise any more money.” The Carlyle Group is the major investor.
“One of the things that drove Renee and a lot of us to join the company was just the unique time in the market that we’re in right now,” Taylor said, adding that this is a once-in-a-lifetime opportunity for semiconductor startups.
“Intel has fallen down both on the design side as well as on the manufacturing side and because of that, they no longer have the advantage they once did,” he said. “In parallel, a company like Arm is finally producing real, server-class IP. And the third thing is this whole hyperscale cloud evolution, which has made it such that the software hurdle to go and move your applications onto an alternative architecture is significantly less than what it was even just a few years ago. A combination of all of those things really created an opportunity for someone like us to come in and disrupt this industry.”