Source: anandtech, theverge, etc.
Editor: Jin Lei, Peng Fei
New wisdom report
【新智元导读】Recently, ARM released the next-generation architecture design solution, three heavy weapon vacated: Cortex-A77 CPU, Mali-G77 GPU and ARM ML processor, the performance is improved. According to ARM, the new flagship IP will define high-end smartphone performance in 2020, providing a new generation of artificial intelligence experience.
ARM offered three weapons.
Yesterday, the world's leading semiconductor intellectual property (IP) providerARMHeavy releaseNext generation architecture design, including:
ARM ML processor.
According to ARM, the new flagship IP will define high-end smartphone performance in 2020, providing a new generation of artificial intelligence experience.
Prior to this, Xinzhiyuan also reported that ARM announced the suspension of cooperation with Huawei, perhaps the flagship IP released by ARM will not be with Huawei.
However, Huawei has now obtained a permanent license for the ARM8 architecture (ARM8 is ARM's 32/64-bit instruction set, and current processors are the product of this instruction set). It is also possible to design the ARM processor completely independently, master the core technology and complete intellectual property, and have the ability to independently develop the ARM processor.
Codenamed Deimos, next-generation CPU performance is greatly improved
The Cortex-A76 released by ARM last year is actually a very successful core architecture. The combination of the new architecture and the 7nm TSMC process brings great performance and efficiency leap. The Kirin 980 and the Opteron 855 are good examples.
But ARM's real competitors are mainly from Apple's CPU design team. As far as CPU is concerned, Apple's A11 and A12 still have obvious performance and efficiency advantages.
The Cortex-A77 code name released by ARM is Deimos (Demos, God of Fear), which is in line with the previous generation A76 and largely maintains the characteristics of the previous generation.
According to the parameters given by the government, there are several key characteristics as follows:
Armv8.2 architecture is still used to support AArch32 and Arch64.
64KB L1 instruction and data cache;
256 or 512KB L2 cache;
And up to 4Mb L3 cache
ARM indicates that Cortex-A77 and Cortex-A76 maintain the same 3.GHz peak frequency target. But manufacturers are not expected to reach such a high frequency in the next generation of Soc.
Cortex-A77 uses more computing power for device security edge computing. These computing capabilities include AI cameras, visual scene detection, 3D scanning, biometric user ID (face recognition), speech recognition, ML in games and ML in AR.
From a performance point of view, Cortex-A77 has many performance improvements over A76:
Single thread performance improves by 20%.
FP performance improved by 35%.
Memory broadband increased by 15%.
Higher performance also means better responsiveness to new applications and experiences that support AR. Through Cortex-A77, augmented reality technology in mobile games is expected to grow substantially in the next few years. Cortex-A77 will also support 5G. Ericsson's mobile report in November 2018 predicts that by 2024, about 1.5 billion smartphone devices will have 5G capabilities.
As we all know, 5G is a basic requirement for computing intensive ML, AR and other emerging devices. It will bring faster speed, wide bandwidth (5 to 20 Gbps), 8K resolution streaming media and 360 degree video.
ARM GPU uses a new architecture, Valhall, with great performance leap forward
In addition to Cortex-A77, ARM has also launched a new generation of Mali-G77.
Key points:Mali-G77 uses a new architecture, Valhall。 The architecture used in G71-G76 is Bifrost.
Mali-G77 has made some significant improvements in performance, density and efficiency. Although last year's Mui-G76 introduced some major changes to the execution engine computing architecture, the G77 has improved even more and is out of touch with ARM's relatively unusual computing core design.
According to the official claim of ARM, Mali-G77 is better than its predecessors.
Efficiency, performance and machine learning performance are improved by 30%, 30% and 60% respectively.
Under the same process and performance, 30% of the energy efficiency improvement is achieved.
It also saves 50% of the power consumption compared with Mali-G72.
According to the official performance, the performance of ARM Mali-G77 is close to that of Apple A12, but also surpasses that of Qualcomm Adreno 640.
ARM ML NPU
As smartphones and computers continue to use AI to realize the functions of scene recognition in photography or object recognition in augmented reality. At the same time, many companies have built specialized neuroprocessing units to deal with these tasks, thus reducing the pressure on CPU and GPU.
This time, ARM also released its latest NPU, providing up to twice the energy efficiency and three times the memory compression of the previous generation. Combining with the open source ARM NN software framework, ARM indicates that the eight-core version of the chip can provide up to 32 TOP/s performance.
What do netizens think?
World VoIP CenterUnfortunately, Huawei can't use it.
AbufrejovalIt is pointed out that while extending all Prefetch (preloading) seems to be a good thing for performance, it also increases the probability of bypass attacks.
SyxbitComplain: every Q of ARM
RaptormissleSo it looks like the SD 865 will eventually break the 4000 geekbench single-core score and probably get 12000 in multiple cores. Arm finally has a place in Apple's competitive environment because I really don't want to see A13 so hot!
GC2:CSI'm very dissatisfied with Apple's fever problem: 76 has made a leap in performance. To be honest, I'd be happy to get the same performance at a lower power. And cut/limit peak power... instead of buying fans and mobile phones
0ironThe evil enchantment sends out the soul with a smile and asks: then A77 belongs to
WilsonkfRespond0ironIt's very possible, but maybe it's too late. Most of the design phase should have been completed long before the ban.
LodixIt seems a bit frustrating to know about Hess's launch of Kirin SOC using ARM IP later this year and all that's happening.
Wilco1Because the performance gap between large kernels is accelerating tearing up, it seems more important to improve the performance of small kernels than to further increase performance.
So dear Xin Zhiyuan netizens, what do you think? Leave a message to tell you what you think of ARM.
Attachment: A Glance of Cortex and Mali's Family
Cortex Family Oriented to Embedded System
ARM processor is widely used in the field of mobile communication because of its low energy consumption, low cost and high performance. Supercomputers, on the other hand, consume a lot of power, and ARM is also seen as a more efficient option.
But the integrated performance of embedded microprocessors on mobile devices is getting higher and higher. In order to meet the increasing demand for computing power, ARM has launched the Cortex series, which represents the cerebral cortex in English.
For different application requirements, ARM has introduced Cortex-A, Cortex-R and Cortex-M. Cortex-R pursues real-time while Cortex-M pursues low cost.
Cortex-A series pursues extreme performance, is the king of performance in Cortex series, and the production industry is constantly challenging the limit. Starting with A 76's 10nm, to A 77's 7 nm, Series A Series will launch 5nm chips in the future.
Series A utilizes multi-core technology to provide extensibility of single to four cores for traditional MPCore processors and up to eight cores for DynamIQ processors. It improves data security and privacy, advanced security and RAS functions, and integrated A I functions, so that DynamIQ-based systems can provide a wide range of solutions. Advanced power management functions integrated in DynamIQ processors can provide higher power savings for energy-saving design. Mainly used in performance-intensive similar, such as mobile devices, autopilot and medical treatment.
Below is a family photo of Cortex-A family.
Main characteristics of Cortex-A series devices:
Extensible Clusters Supporting Single-Core and Multi-Core Configuration
The RISC kernel supports Armv7-A and Armv8-A architectures
Fully backward compatible with previous ARM processor code
VFP and Neon Units Execute Floating Point and Advanced SIMD Instruction Set
Optional encryption accelerator engine, supporting AES, SHA1 and SHA 2-256 algorithms
Memory Management Support (MMU) supports virtual and physical address spaces with various page sizes
Hardware Conversion Table Towards Virtual to Physical Address Conversion
Big-endian and little-endian data access support
Unaligned access support for basic load/store instructions
GPU of ARM: Mali Family
In addition to ARM processor authorization, ARM has a complete set of GPU authorization schemes called Mali. Before, many chip suppliers who did not have independent development of GPU technology directly used the design of ARM processor Mali GPU, such as Samsung, Ruixin micro, exhibition, Italian semiconductor, full ambition and so on.
Mali's family photo:
ARM Mali Series Multimedia Processors provide graphics and display solutions for SoC. It can provide a very good visual experience for high-performance smartphones, tablets based on Android OS and smart TV, which extend from entry-level mass market smartphones to visual shock.
Like other three-dimensional display chips based on IP cores embedded technology, Mali display chipset does not provide a display controller (similar to graphics card) specifically used to drive LCD display images. Instead, it is a pure three-dimensional display engine, which loads images into the cache and displays them by a built-in display core specially responsible for image display processing. These images.