Recently, Ampere announced significant updates to its product roadmap, unveiling the highly anticipated flagship processor, AmpereOne Aurora. This processor features 512 Ampere cores and integrates AI acceleration and High Bandwidth Memory (HBM) on-chip, aiming to significantly enhance the performance and efficiency of cloud-native AI computing.
According to Ampere, compared to existing AmpereOne processors, Aurora is expected to triple the performance per rack. Designed specifically for AI training and inference workloads, this processor demonstrates Ampere's commitment to being a major player in the AI computing field.
Image source note: The image was generated by AI, provided by the image licensing service Midjourney
Aurora employs a scalable AmpereOne Mesh architecture, allowing seamless connectivity for all types of computing, and for the first time integrates Ampere AI IP with high-bandwidth memory into silicon hardware. This design enables Aurora to handle complex and diverse AI workloads, providing robust performance and flexibility for cloud-native computing environments.
Notably, with the release of Aurora, Ampere positions itself as a strong competitor in the high-core-count processor market, directly competing with industry giants like AMD and Intel. Ampere's focus on AI capabilities and cloud-native design gives it a unique selling point in the rapidly evolving data center and cloud computing sectors.
Regarding the process node, although Ampere has not disclosed specific details about Aurora, it has indicated the use of a scalable mesh architecture with multiple chips. Industry insiders speculate that Ampere may collaborate with Intel, utilizing its 18A process or packaging technology to produce Aurora.
Additionally, Ampere introduced other updates to the AmpereOne product line. The current AmpereOne processors feature up to 192 cores and use 5-nanometer technology. By the fourth quarter of 2024, Ampere plans to launch the AmpereOne M series, which will introduce a new 12-channel DDR5 platform, significantly enhancing memory bandwidth. By 2025, Ampere intends to increase the core count to 256, further boosting the processor's performance and computational capabilities.