Intel Arrow Lake-S Appears in Instruction Set Reference Guide
Intel just updated its “Architecture Instruction Set Extensions and Future Features” last month, confirming which instructions will be supported in the upcoming processors.
The guide confirms that both series support a variety of instructions, including AVX-VNNI-INT16, SHA512, SM3, and SM4, which focuses on AI workloads and hash functions.
The arrival of AVX-VNNI on Arrow Lake and Lunar Lake CPUs is especially noteworthy. This instruction set extension improves performance for neural network inference workloads by providing specialized functionality for 8-bit and 16-bit integer arithmetic.
In short, applications that use artificial intelligence, machine learning, and deep learning algorithms can expect improvements in processing speed and efficiency.
The Arrow/Lunar Lake family also supports Intel Linear Address Masking (LAM) instructions alongside Sierra Forest and Grand Ridge.
This allows software to use the untranslated address bits of a 64-bit linear address for metadata.
Conveniently, to make the document easier to read, Intel has highlighted the changes in purple:
* Click the image to open in another window/tab.
Additionally, support for SHA512, SM3, and SM4 instructions on these new CPU series means Intel will improve the security and cryptographic capabilities of its next-generation client CPU series.
The inclusion of SHA512, a widely used cryptographic hash function, strengthens data integrity and enables secure data transfer.
Additionally, this document describes support for the SM3 and SM4 cryptographic hash algorithms used in communications.
Users can also find new hints for the upcoming Raptor Lake Refresh series. The company has strongly backed its plans to add new CPUIDs to future Raptor Lake series and refresh the architecture for desktops and high-end notebooks.
The Refresh is currently said to be part of the 14th Generation Core series rollout that will be offered alongside the Meteor Lake series.
* Click the image to open in another window/tab.
Commentary:
It seems that crypto-related and AI/ML-related instructions will be added to Arrow Lake and Lunar Lake.
I thought, “Will Intel add its own instructions again?”, but it seems that the circumstances are a little different this time.
Aside from encryption-related issues, GPUs are at the core, but now all PC parts are being optimized for AI/ML-related issues at a tremendous pace.
Although only the GPU performance is spotlighted, Phoenix also has an AI-related circuit called AIE.
I can’t even imagine what the world would be like if we could unify the execution environment using CPUs and GPUs, or if AI hardware accelerators for CPUs and GPUs could operate in a mutually complementary manner at the same time. It would be easier and more convenient.
