Recently, a paper in the latest issue of the IEEE Transactions on Cloud Computing pointed out that Alibaba's Yitian 710, released in 2021, is the most powerful and energy-efficient Arm architecture cloud service processor available today.
According to information released by Alibaba, the Yitian 710 is designed for high-concurrency, high-performance, and high-energy-efficiency requirements in cloud scenarios. It adopts advanced process technology, with a single chip accommodating up to 600 billion transistors.
In terms of chip architecture, it is based on Arm's latest ARMv9 architecture, with 128 CPU cores, each core having 1MB of L2 cache, sharing a 64MB L3 cache, and a maximum clock speed of 3.2GHz, balancing performance and power consumption simultaneously.
In addition, to address the bandwidth bottleneck under high concurrency conditions in cloud computing, Yitian 710 integrates the industry's most advanced technologies in memory and interfaces, such as DDR5 and PCIe 5.0 (supporting eight-channel DDR5 memory and 96 PCIe 5.0 lanes), effectively boosting the chip's transmission speed. At the same time, Alibaba has specially optimized on-chip interconnects, utilizing a new flow control algorithm to effectively alleviate system congestion, thereby enhancing system efficiency and scalability, and adaptable to various cloud application scenarios.
Alibaba's disclosed test results at the time showed that Yitian 710 scored 440 points in the SPECint 2017 test, surpassing the industry benchmark by 20%, while also achieving 50% better energy efficiency than the industry benchmark.
For comparison, in the 2021 SPECint 2017 performance ranking, Yitian 710's score of 440 places it in the fifth position on the list, with most higher-scoring chips being high-end server chips based on the x86 architecture from AMD.
When compared to high-end Arm architecture server chips of the same class at the time, Yitian 710's performance advantage is even more pronounced. For example, Kunpeng 920 (dual 128-core, 2.6GHz) scored 318 in SPECint 2017.
In the latest round of testing, "Transactions on Cloud Computing" compared several competitors, including the Intel Xeon Platinum 8848C (32C/3.2GHz/105MB), Kunpeng 920 (60C/2.4GHz/32MB), Ampere Altra (80C/3GHz/32MB), and Amazon Graviton 2/3 (64C/2.5/2.6GHz/32MB). Intel uses the x86 instruction set, while the others utilize the Armv8 instruction set.
In the comparison of eight tests including Dhrystone benchmark, Whetstone benchmark, kernel system calls, number of execl system calls, file copy speed, UnixBench overall score, RSA 2048 sign and RSA 2048 verify, among the same Arm architecture server CPU competitors, Alibaba Yitian 710 is almost "far ahead" in all aspects, with the only average performance shown in Excel.
Although AWS's Graviton 3 outperformed Yitian 710 when running Redis, Yitian 710 emerged victorious in all three workloads measuring memcached throughput and two of the workloads testing RocksDB throughput.
Even when compared to the x86 architecture Intel Xeon Platinum 8848C, it also holds its ground well, performing at a comparable level in both the Dhrystone and Whetstone benchmarks.
The paper also points out that in addition to its outstanding performance, the Ali Yitian 710 features a leading instruction set, DDR5 memory, PCIe 5.0 channels, and a larger L3 cache, all achieved in 2021, which is quite rare.
At the 2022 Yunqi Conference in November 2022, Zhang Jianfeng, the then President of Alibaba Cloud Intelligence, announced that the Yitian 710 had been successfully deployed on a large scale, becoming China's first domestically developed CPU for large-scale cloud applications.
In the future, Alibaba Cloud will continue to expand the deployment scale of its self-developed CPUs, with 20% of the new computing power in the next two years using self-developed chips.
