Ampere and Qualcomm aren’t the most obvious of partners. Both, after all, offer Arm-based chips for running data center servers (though Qualcomm’s largest market remains mobile). But as the two companies announced today, they are now combining forces to offer an AI-focused server that uses Ampere’s CPUs and Qualcomm’s Cloud AI 100 Ultra AI inferencing chips for running — not training — models.
Like every other chip manufacturer, Ampere is looking to profit from the AI boom. The company’s focus, however, has always been on fast and power-efficient server chips, so while it can use the Arm IP to add some of these features to its chips, it’s not necessarily a core competency. That’s why Ampere decided to work with Qualcomm (and SuperMicro to integrate the two solutions), Arm CTO Jeff Wittich tells me.
“The idea here is that while I’ll show you some great performance for Ampere CPUs running AI inferencing on just the CPUs, if you want to scale out to even bigger models — multi-100 billion parameter models, for instance — just like all the other workloads, AI isn’t one size fits all,” Wittich told TechCrunch. “We’ve been working with Qualcomm on this solution, combining our super efficient Ampere CPUs to do a lot of the general purpose tasks that you’re running in conjunction with inferencing, and then using their really efficient cards, we’ve got a server-level solution.”
As for partnering with Qualcomm, Wittich said that Ampere wanted to put together best-of-breed solutions.
“[R]eally good collaboration that we’ve had with Qualcomm here,” he said. “This is one of the things that we’ve been working on, I think we share a lot of really similar interests, which is why I think that this is really compelling. They’re building really, really efficient solutions and a lot of different parts of the market. We’re building really, really efficient solutions on the server CPU side.”
The Qualcomm partnership is part of Ampere’s annual roadmap update. Part of that roadmap is the new 256-core AmpereOne chip, built using a modern 3nm process. Those new chips are not quite generally available yet, but Wittich says they are ready at the fab and should roll out later this year.
On top of the additional cores, the defining feature of this new generation of AmpereOne chips is the 12-channel DDR5 RAM, which allows Ampere’s data center customers to better tune their users’ memory access according to their needs.