Ampere launches data centre chip with 192 custom ARM cores, DDR5
Ampere Computing has launched its family of data centre processors with up to 192 single threaded cores which is currently the highest core count in the industry.
This is the first product from Ampere based on the company’s new custom core built from the ground up with its internal IP. The Instruction Set compatibility is for ARMv8.6+.
The 5nm chips have larger 2MB cloud-optimized private caches, and new cloud features aimed at applications such as AI. These include Mesh Congestion Management, Fine Grained Power Management and Memory Tagging to improve performance consistency, manageability and security for high performance, multi-user environments. This provides predictable performance by mitigating the noisy neighbour challenges in multi-core processing architectures while providing a very secure microarchitecture for multi-tenant cloud environments.
The platform, built at TSMC, uses eight channels of DDR5 memory and 128 lanes of PCI Express Gen5 I/O on devices from 136 cores up to 192 cores operating at consistent and predictable frequencies of up to 3.0 GHz. Each core is single-threaded by design with its own 16 KB L1 Instruction-cache, 64 KB L1 Data-cache, while doubling the L2 private cache to 2 MB.
A coherent mesh interconnect provides efficient bandwidth with 64 distributed home nodes and directory-based snoop filters to enable seamless connectivity between the cores. The mesh congestion management and QOS enforcement features further boost predicable behaviour of data movement with the high core counts.
“Every few decades of compute there has emerged a driving application or use of performance that sets a new bar of what is required of performance,” said Renée James, CEO and founder of Ampere. “The current driving uses are AI and connected everything combined with our continued use and desire for streaming media. We cannot continue to use power as a proxy for performance in the data center. At Ampere, we design our products to maximize performance at a sustainable power, so we can continue to drive the future of the industry.”
Ampere has already shipped its Altra and Altra Max ARM-based processors to Google Cloud, Microsoft Azure, Oracle Cloud, Alibaba and Tencent, as well as leading OEMs like HPE and Supermicro. It says it is ‘intensely focused on fulfilling a commitment to deliver a predictable and rapid cadence of products that lead the market in performance and efficiency’.
“AmpereOne is about more. More cores, more IO, more memory, more performance, more cloud features,” said Jeff Wittich, Chief Product Officer at Ampere. “With our Ampere Custom Cloud Native Cores, this is the next step in the break from the constraints of legacy compute. No other CPU comes close. It is about cloud scale with the maximum performance per rack.”
“The result of all this innovation is the most efficient and highest performance core in the industry for cloud workloads,” Wittich said. “AmpereOne provides Ampere’s customers the highest overall performance, scalability and density for Cloud Native workloads,” he said, adding that Ampere’s Cloud Native Processors offer a performance advantage of greater than 2X over competing CPUs for AI inference due to their lower latency and higher throughput.