Ampere and Qualcomm may not be the most expected partners, given that both companies produce Arm-based chips for data center servers, although Qualcomm’s primary market remains in mobile technology. Nevertheless, the two companies announced today their collaboration on an AI-centric server solution, which integrates Ampere’s CPUs with Qualcomm’s Cloud AI100 Ultra inferencing chips, designed specifically for running AI models rather than training them.
Ampere, like many other chip manufacturers, is keen to capitalize on the AI surge. The company has consistently focused on delivering fast and power-efficient server processors. While the Arm architecture allows it to enhance these features, AI specialization isn’t its primary strength. Consequently, Ampere opted to partner with Qualcomm, with SuperMicro assisting in integrating both technologies, according to Arm CTO Jeff Wittich.
“As I’ll demonstrate impressive performance metrics for Ampere CPUs handling AI inferencing alone, scaling to larger models, such as those with multiple-100 billion parameters, demands a more tailored approach—AI isn’t a one-size-fits-all scenario,” Wittich explained to TechCrunch. “Our collaboration with Qualcomm allows us to combine our highly efficient Ampere CPUs for general-purpose tasks alongside their efficient inferencing accelerators, providing a comprehensive server-level solution.”
In discussing the collaboration with Qualcomm, Wittich emphasized Ampere’s commitment to developing top-tier solutions.
“We’ve had an excellent partnership with Qualcomm,” Wittich noted. “Our collaboration has been very fruitful because we share similar goals, making this partnership particularly promising. Qualcomm is known for their highly efficient solutions across various market segments, and we are focused on delivering equally efficient solutions for server CPUs.”
This strategic partnership with Qualcomm is a key component of Ampere’s annual roadmap update, which includes the introduction of the new 256-core AmpereOne chip. This chip is fabricated using an advanced 3nm process. While these new chips are not yet widely available, Wittich assures that they are ready at the manufacturing level and are expected to be launched later this year.
A standout feature of this latest generation of AmpereOne chips is the 12-channel DDR5 RAM. This enhancement will enable Ampere’s data center customers to better optimize memory access according to specific requirements.
The key selling points for these chips extend beyond just performance. Energy efficiency and operational costs are significant factors, particularly in AI inferencing. Ampere frequently highlights its performance metrics in comparison to Nvidia’s A10 GPUs to underscore these advantages.
Ampere has clarified that none of its current chips will be discontinued in favor of the newly introduced ones. Wittich emphasized that even the older chip models continue to offer significant utility.
Additionally, Ampere unveiled a new collaboration today with NETINT. This partnership aims to create an integrated solution that combines Ampere’s CPUs with NETINT’s video processing technology. The resulting server will be capable of simultaneously transcoding 360 live video channels and adding subtitles to 40 streams using OpenAI’s Whisper speech-to-text model.
“Our journey began six years ago, driven by the belief that this was the right direction,” said Ampere CEO Renee James during today’s announcement. “In the past, low power consumption was often equated with low performance. However, Ampere has demonstrated that this isn’t the case. We have led the way in computing efficiency, surpassing the performance of traditional CPUs within a more efficient operational framework.”
