Recently announcedNew Mali-D77 display processorAfter enabling more stable virtual vision, Arm further announced the launch of the new Cortex-A2019 CPU architecture design and the new Mali-G77 GPU at the Computex 77 pre-show event. It also emphasized that it will expand its provision of machine learning processor designs, allowing mobile devices to achieve 5 teraflops of computing power per watt, thereby improving edge computing efficiency.
As 5G networks continue to grow, Arm believes there will be a demand for higher smartphone performance in the future to accommodate more diverse computing modes. This will also drive more on-device AI computing applications and even provide sufficient performance for more virtual reality, augmented reality, and high-definition streaming applications.
The company announced the launch of the new Cortex-A77 CPU architecture design, which delivers approximately 76% performance improvement over the previously released Cortex-A20 architecture design. The design also incorporates an independent NPU specifically designed for AI applications directly into the complete processor architecture design, enabling processors using the Cortex-A77 CPU architecture to significantly improve performance in applications like machine learning, delivering approximately 35 times the AI computing performance of processors built over the past two generations.
Compared to the previous Cortex-A77 CPU architecture, the Cortex-A76 CPU architecture adds support for the Arm v8.2 instruction set, doubles the branch prediction bandwidth, and improves prediction accuracy. It also improves the branch target cache, increasing it by 1 times in L4 and 2 times in L2 respectively. In addition, the out-of-order core computing bandwidth is increased, and the integer execution bandwidth is increased by 50%. The memory subsystem performance is further improved, and it is compatible with the Arm DynamIQ operating mode design.
ㄠ
In addition, the newly launched Mali-G77 GPU improves performance by approximately 76% compared to the previously launched Mali-G40 GPU. It also adopts the Valhall architecture design and significantly improves the rendering engine, texture processing flow, and cache access efficiency, resulting in an approximately 30% increase in performance per watt.
Using the Mali-G77 GPU to accelerate machine learning operations can improve efficiency by approximately 60%, and increase the efficiency of inference and neural network execution in artificial intelligence computing, thereby making the display performance on the device side smoother.
At the Computex 2019 pre-show event, Arm also stated that it will continue to deepen its collaboration with MediaTek to promote the development of AI technology applications and 5G network computing applications in the future. At the same time, it will also work with SenseTime and Megvii in the field of AI technology applications to create greater growth opportunities for mobile devices.
By the previously proposedProject TrilliumWith the design, Arm has begun to introduce artificial intelligence computing into mobile devices and is compatible with the Android NN software framework, allowing mobile devices to achieve 5 teraflops per watt and 32 teraflops of computing performance, while also doubling power efficiency and tripling memory compression, thereby improving edge computing benefits.
According to Arm's plan, by leveraging the Cortex-A77 CPU architecture, Mali-G77 GPU, Arm ML processor, Arm NN software framework, and the recently launched Mali-D77 display processor, terminal devices will be able to achieve higher computing performance and meet the higher computing performance demands of the 5G era. It can also achieve more comprehensive computing benefits through constant connectivity and cloud collaboration.
