![Jenny Han Dropped a Major ‘The Summer I Turned Pretty’ Easter Egg Revealing [SPOILER] – yahoo.com](https://earth-news.info/wp-content/uploads/2025/09/319001-jenny-han-dropped-a-major-the-summer-i-turned-pretty-easter-egg-revealing-spoiler-yahoo-com-360x180.jpg)
Arm has confirmed that it will be offering its next-gen Arm compute platform, called Arm CSS for Client, at Android smartphones. Executives also mentioned that they could be used for PCs as well.
The announcement follows an earlier report that indicated that Arm might expand its traditional business model. Arm has traditionally sold CPU designs, not silicon, to partners like Qualcomm. Those companies have the freedom to adjust Arm’s designs — depending upon their license agreement — and then ask foundries like TSMC to actually manufacture the chip.
Arm representatives said that Arm will offer “validated and production-ready implementations of the new Arm CPUs and GPUs,” basically maintaining its existing model. (An earlier version of this story said Arm would be selling its own Arm processors fabricated at foundry customers, which is not the case.)
What’s new here is twofold: First, Arm is extending its services to include working with foundrieson 3nm chip designs first focused on smartphones and AI. Second, Arm is launching its next client platform, CSS for Client, led by the Cortex X925 extreme core, the Cortex A725 performance core, the Cortex A520 efficiency core, and the Immortalis G925 graphics engine with ray tracing. Arm is claiming that the new cores will offer the most performance yet, with a 36 percent boost in the Geekbench SC benchmark, running on Android.
“We’re making it easier to build and deploy Arm-based solution and leaving nothing to chance, enabling new performance points, compute capabilities, and helping speed time to market,” said Chris Bergey, senior vice president and general manager of the client line of business at Arm.
There’s a twist, though. Arm is not announcing an AI core, like Qualcomm’s Hexagon. Instead, it’s providing software libraries, called KleidiAI, to target the CPU for AI workloads. Normally, the NPU is seen as the most efficient way of performing AI. According to James McNiven, Arm’s vice president of product management for client, the software will be enhanced for “new features that we’ll be building into the CPUs going forward.”
Phones first, PCs second
Arm is targeting Android phones first, but executives implied that PCs are next up.
“The CPU is a natural first target for AI on Android devices,” Bergey said. “It’s performance and pervasive, and for 70 percent of third-party Android ML [machine-learning] workloads, that’s where they stay.”
Arm, of course, has been welcomed back into the limelight with the new Copilot+ PCs Microsoft just launched with its partner Qualcomm, whose Snapdragon X Elite and Plus chips power them. Nvidia and Arm have also been said to be working together, with reports surfacing last week that the two may be targeting 2025 for a launch. Arm CEO Rene Haas is scheduled to deliver a keynote address at the Computex show in Taiwan next week, where he may offer more details.
What is Arm’s new CSS for Client?
Not surprisingly, Arm says that the new CSS for Client is designed with AI, specifically generative AI, in mind. It’s an Arm V9 architecture, which was launched in 2021 with AI-specific instructions.
Arm isn’t telling us how fast its new CSS platform is, which means that most of the questions an enthusiast would have can’t be answered yet. When I asked about how many TOPS of AI power the CPU and its AI libraries would produce, company representatives demurred. (Arm CSS stands for Arm Neoverse Compute Subsystems, and Arm debuted the first chips for the platform for the data center late last summer.)
“TOPS is a high-level metric that is dependent on frequency, IP choices, and configuration,” the company followed up in an emailed statement. “We are confident that the capabilities offered by our CPUs and GPUs, when complemented with an appropriate frequency and configuration (e.g CSS for Client), can easily achieve the needs of AI PCs and next-gen AI smartphones.”
James McNiven, Arm’s vice president of product management for client, said that the X925 name was changed to show how different that the X925 was from its predecessors. (The 36 percent improvement in Geekbench is a combination of the instructions-per-clock (IPC) improvement and a faster, undisclosed frequency, McNiven said.) It has a wider design, with larger (3MB) private level-2 caches to increase the TOPS count by 50 percent over its predecessor. The time to generate a response in the tiny-Llama AI metric has become faster, with a 41 percent improvement.
The platform’s Immortalis G925 graphics engine “delivers 37 percent more performance on a wide range of graphics applications,” McNiven said.
All told, Arm is promising that the platform will perform web browsing 23 percent faster on Android (via Speedometer 2.1), process YouTube 40 percent more efficiently through software-optimized AV1 video decode, and consume 25 percent less power while gaming. Ray-tracing performance should increase by 52 percent.
The Cortex A725 will be 35 percent more efficient than the A720, McNiven said, and the Cortex A520 efficiency processor will consume 30 percent less power while gaming.
So will these new CSS for Client cores be used in laptops? “We cannot comment on specific customer engagements, and the mobile cadence is pretty well known,” Arm’s Bergey said. “Things like laptops will potentially be a different cadences. But I think there’s considerable amount of interest around Windows on Arm and we think that this is a great solution that can be applied to that market. So stay tuned.”
Correction: Arm is not physically building and selling chips, as an earlier version of this story stated. It will continue to work with partners like MediaTek and Samsung.
>>> Read full article>>>
Copyright for syndicated content belongs to the linked Source : PCWorld – https://www.pcworld.com/article/2344938/arm-aims-its-new-cortex-x925-cores-at-ai-with-an-eye-toward-the-pc.html
