Arm unveiled a smartphone platform with AI and frame generation
Arm has introduced CSS for Mobile 2 – a new platform for flagship mobile chips that combines C2-Ultra and C2-Pro CPU cores, GPU Mali G2-Ultra NX GPUs, the SI L2 system interconnect, software, and developer tools.
The main focus is on local AI and neural network-based graphics processing, which enables improved image quality and frame rates without a comparable increase in computational load.
New graphics
The Mali G2-Ultra NX has become the flagship GPU Arm and, according to the company’s definition, the first AI-native graphics processor in the Mali family. Neural accelerators are built directly into shader cores, making AI processing an integral part of the graphics pipeline.
At the same time, Arm did not limit itself to adding neural blocks. In GPU, the execution mechanism has been redesigned – the company calls this the largest change to the Mali instruction set architecture in seven generations. Each warp now has up to twice as many registers available, giving complex shaders more resources for execution.
According to Arm’s tests, the Mali G2-Ultra NX shows up to 24% higher performance in benchmarks compared to the previous generation. In games without neural network technology, the gain reaches 14%.
AI footage
The main feature of the new GPU is revealed when using neural network algorithms. Arm has prepared three technologies for different stages of image generation. Neural Super Sampling (NSS) reconstructs a high-resolution image from an original frame with lower resolution. Neural Frame Rate Upscaling (NFRU) creates intermediate frames, increasing the final display refresh rate. Neural Super Sampling and Denoising (NSSD) combines neural upscaling and denoising in scenes with ray tracing.
In the demo game Neural Dawn, simultaneous use of NFRU and NSSD improved efficiency by up to four times compared to native rendering. Data exchange with external memory was reduced by 70%.
One of the NFRU use cases is converting rendering from 30 frames per second to 60 frames per second by generating intermediate frames. Arm also views the technology as a way to approach 120 frames per second in mobile games without needing to calculate every frame using traditional methods.
Ray tracing
The capabilities of GPU have also been expanded in the area of hardware ray tracing. The Mali G2-Ultra NX features a third-generation Ray Tracing Unit and hardware support for Opacity Micromaps. This latest technology is designed to more efficiently process complex transparent geometry, such as foliage, fabrics, and multi-layered surfaces.
In one test, Arm’s use of Opacity Micromaps increased frame rates by 30% and reduced ray tracing computational load by up to 70%. The new GPU architecture also reduces DRAM access volume in ray tracing tests by up to 13%.
Faster local AI
CSS for Mobile 2 includes not only the new GPU. The CPU cluster comprises C2-Ultra and C2-Pro: the former is designed for maximum single-thread performance, while the latter targets a balance of speed and efficiency. Compared to the previous generation, Arm claims up to a 15% increase in single-thread performance and up to 12% in multi-thread performance. The company also notes a 15% faster web browsing experience and a 12% reduction in app launch time.
For local AI tasks, SME2 capabilities have been expanded. The flagship reference configuration uses two C2-Ultra chips, six C2-Pro chips, and two SME2 modules. Compared to the previous generation, the number of SME2 modules has doubled.
In Arm tests, this CPU cluster delivered up to 1.7x higher performance on selected AI models. In a full scenario encompassing speech recognition, information retrieval, reasoning, app launching, and web browsing, the C2-Ultra with two SME2 blocks completed the task 24% faster than the previous generation system.
First devices
CSS for Mobile 2 is designed for mobile SoC manufacturers and includes not only compute cores but also the system interconnect SI L2, software, and developer tools. The platform coordinates the operation of CPU, GPU, and other compute blocks, while its individual components can be used independently by manufacturers or combined with their own developments.
The first mass-produced device with the new graphics architecture is the Xiaomi 18 Fold, unveiled in China with the proprietary Xring O3 chip and GPU Mali G2-Ultra NX. The same platform is also used in the Xiaomi Pad 9 Pro Max tablet. Wider adoption of the Mali G2-Ultra NX in smartphones and large-screen devices is expected in 2027.
