The Impact of AMRv9 on the Next Decade: A History of ARM Instruction Set Development

Click the above Computer Enthusiasts to follow us

When it comes to ARM, everyone should be very familiar with it. It does not produce chips; it only provides an idea for chip design, including native core IP and the ARMv instruction set.

Among them, the core IP includes the well-known Cortex-A78 and Cortex-X1. Chip manufacturers such as Qualcomm, Samsung, MediaTek, Huawei, and Unisoc directly purchase core IP licenses and develop SoCs based on it. Chip manufacturers with conditions can also make “magic modifications” based on core IP to launch semi-customized core architectures, such as Qualcomm’s Kryo series cores.

Apple, leveraging its scale, funding, and R&D advantages, chooses to purchase ARMv instruction set licenses from ARM and develop its core architecture independently based on it. In addition, Qualcomm’s early Snapdragon 820 and Samsung’s Mongoose core are also self-developed based on ARMv instruction set licenses. However, the cost of self-developed cores has become too high, making it difficult for all chip manufacturers except Apple to continue.

Whether it is Apple’s self-developed core or ARM’s native IP, they all need to use the ARMv instruction set as a foundation. Therefore, the strength of the ARMv instruction set is crucial.

With the advent of the smartphone era represented by iOS and Android systems, the earliest mobile processors were directly built on the ARMv6 instruction set, with representative products such as Qualcomm’s Snapdragon S1 (including MSM7x25), which did not even have a GPU.

ARMv7-A was a significant iterative update. Qualcomm’s early self-developed Scorpion microarchitecture was developed based on the ARMv7-A instruction set, with representative products including QSD8x50 from the Snapdragon S1 era.

ARM also began to vigorously develop native core IP research and licensing starting from the ARMv7-A instruction set, from Cortex-A5 to Cortex-A17, all of these early 32-bit processor cores were built based on ARMv7-A.

To welcome the 64-bit era, ARM launched the ARMv8-A instruction set in 2016, which features scalable vector extensions, designed for high-performance computing and data centers. Representative products using this instruction set include Cortex-A53, Cortex-A57, Cortex-A72, and Cortex-A73 (as well as A32 and A35 in the embedded field).

In 2017, ARM released the ARMv8.2 instruction set, which introduced new fp16 operations and int8 dot instructions, which can significantly accelerate the inference efficiency of deep learning frameworks if optimized properly. All SoC core architectures from Cortex-A55 and Cortex-A75 to the present are developed based on the ARMv8.2 instruction set.

On March 31, ARM officially released the ARMv9 instruction set, which, while compatible with ARMv8, is no longer limited to the mobile/embedded market. In the future, it will focus on new markets such as PC, HPC high-performance computing, and deep learning to meet the global demand for increasingly powerful security, artificial intelligence, and ubiquitous dedicated processing.

According to ARM, ARMv9 is the most important innovation in the past decade and is the foundation for the future 30 billion ARM chips. It is worth mentioning that the ARMv9 architecture is not subject to U.S. export control regulations, which means that Huawei’s HiSilicon can obtain permanent authorization for the v9 architecture.

It is reported that processors developed based on ARMv9 are expected to officially enter commercial use in early 2022. The most concerning question now is whether the next-generation Cortex-A79 and Cortex-X2 core IP from ARM, which will be released in May to June, can catch up with this brand-new instruction set?

Click “Read Original” for more exciting content

Related posts

Leave a Comment Cancel reply