Why is the 'M1' chip developed by Apple so high-performance?
The
Why is Apple's M1 Chip So Fast?
https://erik-engheim.medium.com/why-is-apples-m1-chip-so-fast-3262b158cba2
The 'M1' developed by Apple is called System-on-a-chip (SoC) , which is a CPU that performs general calculations, a GPU that handles graphics-related calculations, an image engine, a machine learning engine, and a ' Secure ' for security. It is a collection of various dedicated chips such as ' Enclave '. Since the processing related to video and image editing is performed on a dedicated chip, the editing work can be completed faster than when processing with only a CPU for general calculation purposes.
Memory is also the reason why 'M1' is so fast. Since the CPU and GPU memory is not shared in the existing system, it is necessary to copy the data from the main memory to the GPU memory when the GPU uses the data, and the bus bandwidth becomes a bottleneck of processing speed. It was. However, since the unified memory installed in the 'M1' integrates the memory of the CPU and GPU, the GPU can directly access the data without copying the data. Even with other dedicated chips, high-speed data processing is possible as with the GPU.
Dedicated chips and unified memory are not the only reasons why M1 is so fast. The single core performance of ' Firestorm ', which is the CPU core installed in the M1, is higher than most Intel CPUs
When parallel computing is performed by the CPU, instructions are converted and stored in a buffer called Reorder Buffer (ROB) . The 'M1' Firestorm core has three times the ROB capacity compared to Intel and AMD CPUs, so it is possible to acquire more instructions from the ROB and process them in parallel. Furthermore, the number of decoders that perform conversion processing when storing instructions in the ROB has doubled, and instructions can be stored in the ROB at high speed.
The 'M1' was able to increase the number of decoders because of the instruction set it adopted. In CPUs that adopted the x86 instruction set of Intel and AMD, the conversion process by the decoder becomes complicated because the instructions are variable length. On the other hand, RISC adopted by 'M1' does not need to analyze the start point of the instruction because the instruction has a fixed length, and the structure of the decoder is simplified. Therefore, the 'M1' chip can be equipped with more decoders.
Looking at the CPU benchmark scores, Firestorm's single-core performance scores higher than AMD's Ryzen 9 5950X. However, if you look at the operating clock frequency, you can see that the Ryzen9 5950X operates at up to 4.9GHz, while the Firestorm core operates at 3.2GHz. In short, Firestorm has the same performance as the Ryzen 9 5950X at low frequencies. Also, as you can see from the fact that the new MacBook Air does not have a fan, the Firestorm core operating at low frequencies is also advantageous in terms of heat generation.
The business model of Intel and AMD is based on mounting various parts such as CPU, graphic board, memory on the motherboard to complete the PC, but the SoC that handles various functions of the PC with one chip If adopted, there will be technical and intellectual property issues related to each part. Even if Intel or AMD develops SoC, there is a problem that the required hardware and software will change depending on the PC configuration. In that respect, Apple designs and develops both software and hardware in-house, which enables smoother SoC development than other companies.
Finally, Enghien said, 'Fortunately for Intel and AMD, Apple uses the M1 only for its products, so PC user demand will not flow to the M1 chip. Users who have spent a lot of money to build a PC environment will not easily move to Apple products, but young users who have not invested much in any platform will adopt Apple products equipped with the 'M1' chip. Being aggressive with Apple, Apple's influence in the PC market will continue to grow. '
Related Posts:
in Hardware, Posted by log1o_hf