AmpereOne Family Product Brief
AmpereOne® CPUs are the newest and most efficient Cloud Native Processors from Ampere with up to 192 high performance cores. Delivering best-in-class performance per rack, AmpereOne systems are ideal for compute environments requiring the highest level of consistent and predictable performance at scale.
AmpereOne compliments the existing Ampere® Altra® products and adheres to the same Cloud Native principles:
AmpereOne delivers more cores, more memory bandwidth, more memory capacity, more IO bandwidth, and more cloud native features.
AmpereOne | |
---|---|
Custom Core Count | 144 cores | 160 cores | 192 cores |
Private Caches | L2: 2MB/core, L1: 16KB instruction cache/core, 64KB data cache/core |
System Level Cache | 64MB 2x128-bit SIMD units |
Memory |
- 8 channel 80-bit Up to DDR5-5200, 2DPC up to 4400. - Up to 16DIMMs and 4 TB/socket - ECC, symbol ECC, DDR5 RAS features |
Connectivity | 128 lanes PCIe 5.0 with 32 controllers (1P) 192 lanes PCIe5.0 with 48 controllers (2P) |
Cloud Features |
- BFloat 16 - Nested Virtualization - Secure Virtualization - Statistical Profiling Execution - Pointer authentication (EPA2) - Coherent mesh interconnect with distributed snoop filtering - Single key memory encryption - Memory QoS/bandwidth allocation - MTE/Memory Tagging Extensions |
Technology and Architecture |
- TSMC 5nm - Arm v8.6+, SBSA Level5 |
Power |
- Up to 400W TDP supportable - Advanced power management |
AmpereOne processors with up to 192 cores operate at consistent operating frequency of up to 3.2GHz. Each core is single threaded by design, allowing high utilization under maximum loads and eliminating the noisy neighbor effect of multithreaded cores. Eight 80-bit DDR5 channels result in improved memory bandwidth and a larger addressable space of up to 4TB of memory/socket, resulting in overall high performance.
AmpereOne processor’s consistent frequency of operation, large L2 cache size of 2MB, and new memory management features combine to give a predictable performance in a multi-core and multi-tenant cloud environment.
In addition, a coherent mesh interconnect provides efficient bandwidth with 64 distributed home nodes and directory-based snoop filters to enable seamless connectivity between cores. Predictable data movement directory in high core count AmpereOne processor is ensured by the mesh congestion management and QoS enforcement features of AmpereOne.
Cloud resources are meant to be used to their fullest for the sake of efficiency, but it is not always possible due to the unpredictable nature of internet traffic. AmpereOne processor makes rightsizing the infrastructure possible so that applications are neither running at fraction of the max capacity, causing Capex increase nor running close to max capacity, and risking SLA violation.
AmpereOne cloud native processors provide consistent incremental performance on a per core basis. Workloads scale linearly as more processes are added to the single-threaded cores. For dense virtual machine implementations, more VMs can be provisioned on an AmpereOne high core count processor than any other compute in the industry.
Datacenter efficiency is effectively measured by performance/rack which includes power consumption, rack density and the physical data center footprint. The AmpereOne Cloud Native Processor leads all server computer products in the performance/rack metric. As a result, enterprise and cloud service providers can benefit from lower power consumption, higher operational efficiency, and lower cost of operations.
AmpereOne, as a dense compute product, allows cloud native workloads to scale out, providing maximum rack capacity, outperforming legacy x86 products at rack level. The resulting efficiency gains help reduce IT Opex in the order of $10M-$100M.
Data center operators and IT designers can now have the advantage of high-density compute environments to meet their unprecedented compute capacity demands. Additionally, a lower carbon footprint with AmpereOne will allow IT enterprise managers to meet aggressive ESG (Environmental, Social and Governance) goals and shorten the time to Net-Zero emission initiatives.
With up to 128 lanes of PCIe Gen5 (32 GT/s per lane) per socket, AmpereOne provides the flexibility to add up to 32 off-chip devices, including networking cards, storage/NVMe devices, GPUs, and other accelerators. AmpereOne based platforms are well-suited for a variety of system configurations delivering the high performance for large capacity storage, networking, AI inference, and the newest generative AI models and applications.
Features | Benefit |
---|---|
Speculative Side Channel Attack Mitigation | Additional protection against a class of potential attacks that use side channels |
Buffer Overflow Protection/Memory Tagging | Prevents a class of potential attacks that use buffer overflow to compromise data |
ROP/JOP Attack Mitigation | Prevents a class of potential attacks that use types of application exploits used in the past to compromise processes in execution |
Crypto and Entropy Acceleration | Improves performance for RNG, SHA512, SHA3 cryptography algorithms |
Single-Key Memory Encryption | Protects memory from being maliciously read by unauthorized actors |
Product Name | Cores | Sustained Frequency (GHz) | TDP* |
---|---|---|---|
A192-32X | 192 | 3.2 | 400 |
A192-26X | 192 | 2.6 | 300 |
A160-32X | 160 | 3.2 | 350 |
A160-28X | 160 | 2.8 | 300 |
A144-28X | 144 | 2.8 | 275 |
A144-24X | 144 | 2.4 | 250 |
A96-27X | 96 | 2.7 | 200 |
*Note that the actual power usage depends on specific operating conditions (workloads, clock frequency etc.) and is different (lower in case of highly power-efficient Ampere Cloud Native Processors) than the specified TDP.
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 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 substantially comply with the latest available datasheet.
Arm is a registered trademark of Arm Limited (or its subsidiaries) in the US and/or elsewhere. All other trademarks are the property of their respective holders.