Former Intel President Renée James has launched a new startup aimed squarely at the nascent ARM server market. James left Intel back in 2015, during a major reorganization, and her new company, Ampere, has been operating in stealth mode. What’s interesting is the design she’s planning to bring to market and the previous platform for the product.
Last year, the wireless communication company Macom purchased Applied Micro Circuits Corp (AMCC) and started shopping for a buyer for that company’s X-Gene SoC family. At the time, Applied had built the X-Gene 3, an ARM server core that wasn’t formally ready for prime time. The IP was sold to the Carlyle Group, a private equity firm. Now, X-Gene 3 is being relaunched as a new product. The design was first announced in 2015, which means the CPU Ampere wants to bring to market in 2018 will be based on a CPU core that’s nearly three years old.
In and of itself, that’s not automatically a disqualifying factor. CPU performance improvements have slowed over time, older chips are still quite effective at handling various workloads, and there’s room in the market for a wide range of capabilities and price points. Just because an architecture is older doesn’t mean it’s automatically worse. And the Carlyle Group is willing to support her efforts, having hired 250 people to work on the project according to EETimes.
Thirty-two cores for $950 with a 3.3GHz clock speed isn’t a bad deal, intrinsically, but there are a huge number of questions about the CPU’s performance, throughput, and flexibility. It’s also not clear if the ARM server market is ready for prime time.
AMD ran into problems on this front several years ago, when it wanted to bring its own Cortex-A57 solutions to market. While this wasn’t the only problem the company faced, one critical issue was that the ARM software stack simply wasn’t ready. A great deal of mission critical validation and code hadn’t been authored, and this harmed AMD’s attempts to bring a chip to market. Three years later, that situation has probably improved — we’ve seen companies like Qualcomm positioning Centriq to take on Intel — but actual purchase orders seem thin on the ground.
By taking on companies like Intel and Qualcomm, Ampere is setting the stage for a tough fight. The pivot to AI and deep learning makes developing a follow-up architecture critical; the workloads of 2015 (when X-Gene 3 was new) don’t bear much resemblance to the workloads of 2019 or 2020. There’s a massive industry-wide pivot towards very different types of workloads underway, and companies that don’t read their tea leaves properly and adjust accordingly are likely to be squashed.