
If it wasn’t already abundantly apparent, neural rendering is the next phase of graphics rendering pipelines. With NVIDIA’s DLSS 5, Apple’s Neural GPU, and Qualcomm’s Adreno Neural Fusion, we are already seeing quite a bit of activity on that front. Arm actually teased that it would be releasing AI-native graphics last year, and here we are with Arm’s first neural GPU architecture. The GPU isn’t the only part of Arm’s CSS For Mobile 2 Platform, with new CPUs and interconnects to bring them together.
Arm CSS for Mobile 2 Brings AI To The Forefront
Arm’s CPUs have long been at the forefront of AI processing because many AI workloads pass through or are managed by the CPU before they ever reach an accelerator. Additionally, with the advent of agentic AI, there’s more demand than ever for CPU cores and cycles to run different agents. That’s why Arm is once again releasing an entirely new CSS platform that leverages the company’s latest CPU and GPU IP with the best AI acceleration it has to offer.
Arm Is Building Agentic CPU Clusters
Building on last year’s C1 family of CPU cores, Arm is releasing the C2-Ultra and C2-Pro CPU cores with twice as many SME2 engines for better SME2 performance. Arm claims 15% single-thread and +12% multi-thread performance improvements measured via FPGA. This is a bit out of character for Arm, as it usually has test chips proving out performance, so we will have to wait for real-world performance to verify these claims. The CPU is without a doubt the center of Arm’s mobile universe, but it isn’t the end-all, be-all, especially with the new GPU.
Mali G2-Ultra NX Brings Neural Graphics to Mobile
Arm’s new GPU is a significant departure from the previous generation architecture and demonstrates probably the biggest leap in capabilities for an Arm core, ever. Arm is introducing three new cores as part of this flagship-class GPU. First, the Neural Accelerator (NX) is tightly integrated into the GPU for sub-1W neural graphics. Second, there is an entirely new execution engine (EE) with twice as many registers, specifically designed to accelerate UE5 features like Nanite and Lumen. Third, Arm also introduced its third-generation Ray Tracing Unit (RTU v3), which adds opacity micromaps and reduces GPU workload by up to 70% during ray tracing. Arm has impressed me with its GPUs ray tracing performance, so it will be great to see how this one turns out.
CSS for Mobile 2 Keep Arm and Its Partners In The Hunt
The CPU is still very much at the center of Arm’s agentic AI universe, but now it also has a powerful GPU core. The Mali G2-Ultra NX GPU is Arm’s most powerful and also most capable of AI acceleration to complement the CPU. That said, there will still be an NPU, even if Arm doesn’t want to talk about it. Mobile platforms are moving towards a balanced AI workload environment, where the best core for the job takes over, improving performance and battery life by maximizing AI compute efficiency. I expect we will see SoCs leveraging these latest cores in devices before the end of the year, possibly even the end of the month.
