Arm's Next-Generation GPU Architecture Almost 200% Faster Than Before

At its DevSummit conference this week, the Arm said that is the construction of the next generation GPU which is supposed to be 2022 that will work almost twice compared to this year's Arm Mali-G710 in terms of FP32 engine performance. Compared to the previous generation of Arms building since 2018, the new GPU will be almost five times faster than the stated load.

Speaking at DevSummit, Ian Bratt of Arm, Chief Technology Officer of the ML business unit at Arm, showed a slide showing how fast the architecture of Arm's GPU architecture is progressing rapidly. The company’s 2022 GPU structure is now expected to be 4.7 times faster than the Mali-G76 at FP32 ML load per performance. The GPU core is a collection of application units, texture units, raster units, and other application-specific computer hardware.

It is not yet clear how the performance enhancement of FP32 ML affects game performance or overall performance per watt, but we do know that the latest Arm Mali G-710 offers 35% higher ML performance and 20% higher image performance in ISO-process GPU code configuration compared to Mali-G78 application. That being said, it is clear how important ML performance is to Arm. At the same time, Arm needs to make sure software developers have the tools to take advantage of its latest technology.

"In addition to adding commands and improving hardware IP, we must also provide software, tools, libraries to run this ML," Bratt said, according to the Register.

Arm has greatly accelerated the development of its GPU architecture in recent years. Back then, the Arm was able to stay in one GPU building for about five years and while those structures appeared purposefully over time both through the performance of each spine and the number of cores supported, these structural performance enhancements were not available. exactly the success. Since 2016, Arm has been in the space of a three-year state of the art GPU while continuing to introduce a new iteration of its architecture every year.

That acceleration has led to a rapid shift in skills and performance. For example, the Mali-G710 (based on the Valhalla 3 architecture) introduced this year is two or three times faster than the Mali-G76 (Bifrost 3 architecture) announced in 2018 regarding the performance of each graphics.

It remains to be seen how the Arm's GPU's architecture will emerge if (at any time) regulators around the world approve the company's acquisition by Nvidia. The company Santa Clara, California is a global provider of various GPUs and computer GPUs for a variety of applications, including machine learning, so it is unclear whether Nvidia continues to develop Mali GPUs, or will redistribute to Mali developers. Its own GPU architecture.