ARM announced new processors Cortex-A75, A55 and Mali-G72





Today, at the Computex 2017 technology conference, ARM unveiled new processors that it plans to bring to market within a year. This is a powerful Cortex-A75 processor and an energy-efficient Cortex-A55 processor, as well as a new Mali-G72 graphics processor.



Cortex-A75 and A55 were the first chips with technology DynamIQ big.LITTLE, which was described earlier . It is specially optimized for AI applications that will become commonplace in future smartphones, for new interfaces (voice control) and virtual reality. Probably, hardware support for AI is also useful for a single distributed AI system that can work on billions of Android smartphones β€” Google developed such a system and is now beginning to be tested .



With hardware support for machine learning and AI, new special processor instructions (for example, calculations with limited accuracy) will be available to developers. New processors will provide a significant increase in performance in AI applications, in comparison with current systems based on Cortex-A73.







ARM big.LITTLE is a heterogeneous computing architecture where space is allocated on the CPU for special hardware accelerators of machine learning applications. A special feature is the presence of two types of processor cores: relatively slow, energy efficient (LITTLE) and relatively powerful and voracious (big). Background tasks on a smartphone or other device are convenient to solve with small cores that consume very little energy. If necessary, the processor activates powerful voracious kernels, which, in multi-threaded mode, working together, demonstrate particularly high performance. In principle, all cores have access to shared memory, so that tasks can be set to run on both types of cores simultaneously. That is, large and small switch on the fly.







DynamIQ big.LITTLE - an evolutionary step forward. The new architecture allows you to use a variety of combinations of large and small cores that were not previously possible. For example, 1 + 3, 2 + 4 or 1 + 7. Up to eight cores can be located in each cluster. Almost unlimited scale DynamIQ clusters with shared memory is a proposal to create powerful computing systems of various purposes.



Cortex-A75 (4 large cores + 4 small)



The Cortex-A75 processor (3 GHz), according to the British company, provides the performance level of a modern laptop, consuming no more power than current smartphones. The processor replaces the Cortex-A73 (2.8 GHz), and according to the SPECint2006 test, its performance is 20-40% higher than its predecessor.















Cortex-A75 is supposed to be used not only in smartphones and tablets, but also in laptops (like Chromebooks), cars and other devices with a large screen - everywhere where high performance is needed.



Cortex-A55 (1 large core + 7 small)



Cortex-A55 is a completely different bird. It is optimized to save energy and consumes 2.5 times less than the current generation of processors Cortex-A53 (performance gain 14-97%). It will be used in the same smartphones and tablets, if high performance is not required, as well as in various devices of the Internet of Things.











The Cortex-A55 (16 nm) will replace the Cortex-A53 (28 nm) - a relatively old CPU, which has become the most popular 64-bit processor in the history of computer technology.



Both processors have native support for HDR and the new CryptoCell-712 security subsystem.



Mali-g72



The new graphics processor Mali-G72 promises 40% more performance than processors and devices in 2017, and 25% better energy efficiency.



The graphics processor is specially designed with the expectation of virtual reality. For example, it supports Multi-View β€” an advanced rendering function that is used in Samsung Gear VR virtual reality helmets, supports Foveated Rendering technique for tracking the look in VR helmets, as well as Adaptive Scalable Texture Compression (ASTC), Multisample Smoothing (MSAA) and smoothing the corners of objects.







This chip is also optimized for AI applications - for example, matrix multiplication (GEMM) is 17% more efficient here, ARM said. It is assumed that the graphics devices Mali-G72 will be used in conjunction with the processor Cortex-A75 or Cortex-A55. According to ARM statistics, last year 50% of smartphones sold and 50% of VR helmets worked on the Mali graphics.



The manufacturers of hardware devices, new chips should appear in I quarter. 2018 year.



All Articles