Arm In-depth丨In order to bring higher quality experience to mobile consumers, Arm is also working hard!

With a new execution engine design, Mali-G52 can achieve higher performance in a smaller chip area, supporting mainstream devices to implement advanced features such as machine learning (ML) and augmented reality (AR); while Mali-G31 is the smallest GPU launched by Arm so far, supporting the latest generation of APIs for digital TV (DTV) graphics development and complex user interfaces. So, how do these two new GPUs help the next generation of new technologies?

You might think that only the most high-end mobile devices are equipped with machine learning features, but surprisingly, everything is developing faster than you expected. Even on mainstream mobile phones, we also expect the device's camera to automatically recognize faces so that it can focus accurately, or to be able to accurately search for photos of pets in the album. You may not realize that these features are all made possible by machine learning.

Arm recognizes the importance of machine learning and recently released the Project Trillium machine learning platform. However, while Arm’s powerful new generation of machine learning processors performs very well for high-end machine learning workloads, this is only one case in a wider range of application scenarios. As we have recently discussed, in mainstream and entry-level devices, it is often impractical to equip such a processor with a sufficiently powerful chip, but for less loaded functions, Arm GPUs or CPUs are usually suitable.

Mali-G52 has once again pushed all performance levels up a notch. Building on the success of the previous generation of GPUs based on the Bifrost architecture, Mali-G52 uses a newly designed execution engine - the part of the processor that executes algorithms, achieving higher performance required for machine learning within a lower chip budget. Compared to the previous generation of processors, Mali-G52 doubles the number of threads per execution engine , doubling the computing performance of all complex content, but the size has only increased by 22%, so partners will not worry about high chip budgets. Not only that, Mali-G52 also introduces support for Int8 dot . Machine learning inference algorithms on terminals widely use general-purpose matrix multipliers, but they usually do not need to reach the precision level of FP16/FP32. In many cases, Int8 is equivalent and more efficient. Mali-G52 has a processing power of four cycles per thread per execution engine, achieving an outstanding performance of nearly 4 times the machine learning performance of the previous generation of processors in current image detection and other machine learning benchmarks.

More immersive content experiences will no longer be exclusive to the highest-end devices. Just because you don't want to wear a VR headset for a full hour (or maybe you don't want to wear any additional equipment at all, but can still enjoy a fully immersive experience through some sophisticated ways), it doesn't mean there are no other options for achieving VR effects. AR augmented reality is quickly becoming the preferred reality technology, and it has become crucial to support this function and features such as 360-degree panoramic video in mainstream devices.

Mali-G52 is the most energy efficient processor in its class, and the intelligent tasking capabilities of the DynamIQ CPU architecture really push the performance envelope. The most popular combination in this generation of mainstream mobile solutions is a Cortex-A75 processor paired with seven Cortex-A55 processors in a DynamIQ configuration, where these smaller cores can handle multiple less cumbersome tasks simultaneously, allowing the powerful Cortex-A75 to focus on processing areas where its overall performance is truly needed. Thanks to these advances, Mali-G52 has a 30% higher performance density than the previous generation , processing highly complex content in an instant. Similar to the flexibility of DynamIQ, Mali-G52 has extremely scalable implementation options, allowing our partners to achieve the exact balance of performance and efficiency required for their devices. Partners can choose two or three execution engines per core, with a maximum of four cores. This additional scalability is a significant advantage for cost-constrained mid-range and low-end devices, as it provides each partner with a very high level of granularity when weighing priorities.

It used to be that if you had less than a certain level of support on your phone, you didn’t think there was a better option than something like Snake. Today, with better displays and more stunning visuals, we’ve raised expectations in terms of user experience. We talked a lot last year about the mobile gaming industry surpassing console and desktop gaming in terms of revenue and appeal, driven by high-fidelity mobile games, and this trend is also extending to mainstream and ultra-efficient endpoints, which is where our second new GPU, Mali-G31, comes in. Designed to meet developer needs, Mali-G31 is Arm’s first ultra-efficient GPU based on the Bifrost architecture . Mali-G31 is the smallest processor to support both OpenGL ES 3.2 and the Khronos API Vulkan , which means developers can access these APIs through a new content performance tier, achieving huge energy efficiencies and optimizations.

With millions of mainstream and entry-level devices now available, app developers can optimize their content to run perfectly on all levels of devices powered by Mali chipsets. This not only improves quality, but also quantity - more games and apps will run on more devices, benefiting more users. It also means developers can access a larger market and continue to develop their apps for the next stage. All links in the mobile ecosystem, from silicon partners to gamers, will benefit from this.

We’ve talked about the present and future of these new technologies, but they are not the only things pushing today’s low-cost systems to the limits of their capabilities. If you’ve recently upgraded to a Smart TV, the new chip technologies are providing a quantum leap in user experience when it comes to user interface and interactivity compared to previous chip technologies. Today, we no longer watch just a set of channels from one content provider, or a set of channels at once. With streaming services like Amazon Prime and Netflix integrated directly into the system, the system must support a multitude of different interfaces, graphic overlays and display composition layers. Add to that the fact that many users now use voice-activated search to select the shows they want to watch, and the seemingly simple digital TV has evolved into something very smart. Mali-G31 will enable the future of digital TV with a 20% smaller size, reduced cost for chip manufacturers, while meeting the workload of user-specific interfaces. When implemented with the other new Mali Multimedia Suite products ( Mali-V52 and Mali-D51 ), the new mainstream solution will provide our customers with significant performance and efficiency improvements and real competitive advantages.

As innovative technologies continue to develop, the development territories of Arm and its partners are expanding, and the mainstream device market is entering areas that were originally exclusive to the high-end market, thereby providing the next generation of visual experiences and fascinating immersive content to new consumer groups.