Ampere Computing announced the details of AmpereOne processors
Last May, following the announcement of breaking free from reliance on the Arm architecture and unveiling the novel, independently architected processor brand AmpereOne, Ampere disclosed details of the inaugural batch of Ampere One series processors in the Chinese market. These include a single-thread CPU design with up to 192 core configurations, which continues to utilize the Arm instruction set, thus ensuring compatibility with the existing Ampere Altra and Ampere Altra Max processors designed with the Arm Neoverse architecture.
The AmpereOne series processors are not intended to replace the existing Ampere Altra series products. Instead, they target design for cloud-scale computing with higher computational density and voluminous data throughput applications, responding to the demands of large-scale cloud computing applications.
Embodying Ampere’s own core designs, distinctions are made between 136-core, 144-core, 160-core, 172-core, and the highest 192-core design. The peak operating frequency can reach 3.0GHz. They are produced with a 5nm process, each core configured with 2MB L2 cache memory and provide a 16KB cache for instruction set applications. Furthermore, they provide a 64KB cache design for data application needs, while the L3 cache memory reaches 64MB.
Regarding memory, they are compatible with 8-channel ECC DDR5 specifications, accommodating up to 16 memory slots and 8TB capacity. They also support up to 128 PCIe Gen 5 channels, or alternatively the use of 64 CCIX channels. They feature network congestion management, power management, and memory tagging functions to respond to the need for faster cloud AI computational demand, thereby optimizing cloud computing.
Ampere accentuates that the AmpereOne series processors deliver higher computational performance for cloud-native workloads while maintaining high scalability and computational density. Furthermore, with low latency and high data throughput capabilities, they enable cloud-based AI inference performance to exceed twice the performance of other brand processor products.
Cloud service providers, including Google Cloud, Microsoft Azure, Oracle, Alibaba, and Tencent, as well as server providers like HPE and Supermicro, are all deeply collaborating with Ampere.