Ampere’s AmpereOne® M processor is part of the AmpereOne® family of high-performance server-class processors. It is engineered to deliver exceptional performance for AI Compute: from inference to the broad spectrum of mainstream data center workloads. With 12 DDR5 memory channels, AmpereOne® M is ideal for computationally intensive generative AI (GenAI) applications such as large language model (LLM) AI inference and memory-intensive object caching. Its high memory bandwidth drives superior performance and maximum model density, empowering clouds and enterprises to handle more AI inference tasks per processor in secure virtual machines or containers with ease.

AmpereOne® M processors feature a new platform architecture with a higher pin count and additional memory channels, distinguishing it from other AmpereOne platforms while continuing to maintain Ampere’s Cloud Native processing principles including:

• Performance: Delivers optimal tokens-per-second performance and maximum instance, session or agent density for advanced AI-enabled digital services
• Efficiency: Minimizes energy consumption for AI Compute workloads to enable sustainable and high capacity data center operations
• Scalability: Seamlessly adapts to evolving AI Compute requirements by supporting the full range of compute workloads within virtualized or containerized environments on a single easy-to-deploy platform

AmpereOne® M implements an advanced and innovative architecture that delivers power-efficient AI Compute at lower cost.

AmpereOne® M processors with 192 custom Ampere Arm ISA-compliant cores deliver consistent high performance and memory bandwidth, while maintaining a TDP that ranges from 250W to 425 W. Each core has a dedicated low-latency 2 MB L2 cache and maintains consistent predictable performance, mitigating noisy neighbor interference. 

With 12 DDR5-5600 memory channels, AmpereOne® M processors provide high memory bandwidth while supporting memory capacities of up to 3 TB, allowing a large number of different AI models to be loaded into memory and executed simultaneously – a capability not possible with GPU-based AI inference implementations. 

Ampere’s coherent mesh topology provides efficient bandwidth with 64 distributed home nodes and directory-based snoop filters to enable seamless core to core connectivity. The architecture also incorporates Adaptive Traffic Management and QoS enforcement, further enhancing predictable data movement. 

AmpereOne® M processors, featuring high computing performance, large memory bandwidth and capacity, as well as flexible interconnect via PCIe Gen5, are optimized to meet the demanding needs of dense AI inference workloads. 

AmpereOne® M establishes new levels of power efficiency for dense cloud and enterprise AI workloads. Its high-density compute design minimizes rack and floor space requirements, reduces overall power consumption, and drives total cost of ownership (TCO) savings. AmpereOne® M processors also utilize the following advanced power management features to enhance efficiency:

• Advanced Configuration and Power Interface (ACPI) v6.4 for sophisticated power control
• Adaptive Voltage Scaling (AVS) for dynamic power optimization
• Dynamic Voltage Frequency Scaling (DVFS) for responsive power-performance tuning
• Fine-grained thermal monitoring and power provisioning
• Dynamic power estimation for real-time power adjustments

These innovations establish AmpereOne® M as the foundation for power-efficient, high-performance AI computing, enabling data centers to achieve optimal efficiency, maximum AI workload density and much lower operational cost compared with legacy GPU-based infrastructure.

AmpereOne® M processors, built from a large number of single-threaded cores with dedicated resources connected by an intelligently controlled high bandwidth mesh, ensure workloads scale linearly with demand. The processors are optimized to maximize rack density for virtualized and containerized deployments of AI Compute, consisting of workloads from inference to general purpose.

With 96 lanes of PCIe Gen5, AmpereOne® M offers maximum flexibility to connect off-chip devices, including networking cards and high-performance storage/NVMe devices. This design makes AmpereOne® M processors an optimal choice for AI inference and high memory-bandwidth applications, enabling organizations to flexibly deploy and scale Cloud Native AI applications.

Reliability, Availability, and Serviceability (RAS)

AmpereOne® M processors delivers robust data center-grade RAS features, ensuring high system integrity and resilience. Key capabilities include:

• Advanced ECC Memory Protection: Data in memory is protected from errors by advanced ECC, maintaining system stability.
• End-to-End Data Poisoning: Corrupted data is tagged and any attempt to use it is flagged as an error, preventing propagation of faulty data.
• SLC ECC Protection: Error correction is ensured for the System Level Cache, maintaining data integrity.
• Background Scrubbing: System Level Cache and DRAM is continually scanned to detect and correct single-bit errors before they escalate into uncorrectable faults.

These advanced RAS features make AmpereOne® M ideal for mission-critical enterprise and AI driven cloud applications, offering enhanced reliability and operational continuity in modern computing environments.

*Usage power data is subject to change based on system configuration and other factors. Usage power is defined as average power consumed over time while executing the SPECrate®2017_int_base benchmark.

The Export Control Classification Number (ECCN) for the AmpereOne® M family products is 5A992.C. For additional information, please contact  tradecompliance@amperecomputing.com.

Ampere Computing reserves the right to change or discontinue this product without notice.

While the information contained herein is believed to be accurate, such information is preliminary and should not be relied upon for accuracy or completeness, and no representations or warranties of accuracy or completeness are made.

The information contained in this document is subject to change or withdrawal at any time without notice and is being provided on an “AS IS” basis without warranty or indemnity of any kind, whether express or implied, including without limitation, the implied warranties of non-infringement, merchantability, or fitness for a particular purpose.

Any products, services, or programs discussed in this document are sold or licensed under Ampere Computing’s standard terms and conditions, copies of which may be obtained from your local Ampere Computing representative. Nothing in this document shall operate as an express or implied license or indemnity under the intellectual property rights of Ampere Computing or third parties.

Without limiting the generality of the foregoing, any performance data contained in this document was determined in a specific or controlled environment and not submitted to any formal Ampere Computing test. Therefore, the results obtained in other operating environments may vary significantly. Under no circumstances will Ampere Computing be liable for any damages whatsoever arising out of or resulting from any use of the document or the information contained herein.

Ampere is not responsible for any losses or damages of any kind incurred because of using or relying on this information. Each purchaser, consignee or user of Ampere products is solely responsible for its compliance with all applicable laws and regulations.

Ampere Computing reserves the right to make changes to its products, its datasheets, or related documentation, without notice and warrants its products solely pursuant to its terms and conditions of sale, only to comply with the latest available datasheet.

©2025 Ampere Computing LLC. All Rights Reserved. Ampere, Ampere Computing, AmpereOne and the Ampere logo are all registered trademarks or trademarks of Ampere Computing LLC or its affiliates. Arm is a registered trademark of Arm Limited (or its subsidiaries) in the US and/or elsewhere. All other product names used in this publication are for identification purposes only and may be trademarks of their respective companies.