Ampere Unveils 3nm 256-Core CPU, Partners with Qualcomm

Ampere Computing, a leading provider of high-performance server processors, has announced its annual strategy update and product roadmap, featuring the revolutionary AmpereOne® platform with a 256-core CPU based on a 3nm process. The company also revealed a strategic partnership with Qualcomm Technologies to develop advanced AI inference solutions.

In a joint announcement, CEO Renee James and Chief Product Officer Jeff Wittich introduced the new AmpereOne® platform. This state-of-the-art platform, leveraging the N3 process node, includes 12 memory channels and a 256-core CPU. The forthcoming 256-core AmpereOne® CPU will maintain the same air cooling solution as the current 192-core version, offering over 40% higher performance than any existing CPU, without the need for complex platform designs. The 192-core, 12-channel memory platform is slated for release later this year.

Ampere emphasized that the AmpereOne® platform continues to set the standard for performance per watt. Compared to two leading x86 processors in the market, AmpereOne® delivers 50% and 15% higher performance per watt, respectively. For data centers aiming to upgrade and consolidate older infrastructure, AmpereOne® promises up to 34% enhanced per-rack performance, resulting in significant space, budget, and power savings.

Performance data shows that AmpereOne® improves per-rack cloud-native performance by 58% compared to AMD Genoa and by 42% compared to AMD Bergamo. Solutions based on AmpereOne® can achieve up to 33% rack space savings, 15% server savings, and 35% power savings over those based on AMD Genoa/Bergamo.

The new AmpereOne® OEM and ODM platforms are expected to ship in the upcoming months.

Ampere also provided updates on Meta's Llama 3 running on Oracle Cloud with Ampere CPUs. Performance metrics indicate that running Llama 3 on a 128-core Ampere® Altra® CPU (without GPU assistance) matches the combined performance of Nvidia A10 GPUs and x86 CPUs, while consuming only one-third of the power.

Ampere is collaborating with Qualcomm Technologies to create solutions using Ampere CPUs and Qualcomm's Cloud AI 100 Ultra, aimed at addressing large language model (LLM) inference challenges in generative AI applications.

Additionally, Ampere announced the formation of a UCIe working group as part of the AI Platform Alliance. This initiative seeks to enhance CPU flexibility by integrating customer IP into future CPUs through open interface technology.

Ampere also introduced a joint solution with NETINT, combining NETINT's Quadra T1U video processing chip and Ampere CPUs. This solution can transcode 360 live channels simultaneously and provide real-time multilingual subtitles for 40 streaming channels using OpenAI's Whisper model.

Alongside existing features such as Memory Tagging, QOS Enforcement, and Mesh Congestion Management, Ampere launched a new FlexSKU feature, enabling customers to manage both horizontal and vertical scaling use cases with the same SKU, further increasing operational flexibility.