In announcing details of the Lumex compute subsystem, designed for the flagship mobile platform, Chris Bergey, Arm Senior Vice President and General Manager of the End Products Group, stated that the subsystem utilizes a new CPU cluster name, abandoning the long-standing Cortex naming scheme. He also anticipates that the new subsystem design will further drive the growth of the Arm design licensing ecosystem.
Abandoning the Cortex naming method is in response to development needs
Chris Bergey explained that when product names were previously adjusted to meet different computing needs to more clearly identify specific computing scenarios, they had already stated that they would use naming principles such as Ultra, Premium, Pro, Nano, and Pico to distinguish performance levels. This made it easier for developers and partners to understand the features and version differences through Arm product names.
Therefore, the company introduced the new Lumex computing subsystem and announced C1 as the name of the new CPU computing cluster. It also introduced the new Mali G1 series name for the GPU, using Ultra, Premium, and Pro as product performance differentiation, discontinuing the previously proposed Cortex name. While the GPU retains the Mali name, it also drops the Immortalis name. The Mali G1-Ultra, Mali G1-Premium, and Mali G1-Pro naming scheme provides more intuitive performance positioning recognition.
James McNiven, vice president of product management at Arm's End Products Division, explained that the C1-Ultra will replace the previous Cortex-X series, the C1-Pro will replace the existing Cortex-A700 series, and the C1-Nano will replace the Cortex-A500 series. The newly launched C1-Premium is a new design, primarily targeted at the design needs of sub-flagship mobile phones, offering similar computing performance while reducing the footprint by up to 35%.
As for the new computing cluster name Pico mentioned earlier, it corresponds to the previously launched Cortex-A300 series CPU, but it itself is not part of the Lumex computing subsystem and is therefore not used in the newly launched C1 series CPU.
As for the core combination of the computing cluster, James McNiven explained that it still follows the previous DynamIQ and big.LITTLE combination rules, which means that different core architectures can be more flexibly constructed to meet actual computing needs, thereby meeting differentiated computing performance.
However, as it is currently in a product transition phase, Arm will retain some Cortex series CPU designs. For example, the Arm Zena computing subsystem, previously designed for automotive applications, utilizes the vehicle-specific Cortex-A720AE CPU high-performance computing cluster. Furthermore, the Arm Cortex-R64AE, a 82-bit processor with safety features, provides high-performance computing for automotive zone controllers and safety islands. However, these are expected to be gradually replaced with newly named products as the product lifecycle evolves.
Differences between SME and SME 2 extended instruction sets
James McNiven pointed out that the difference between the SME and SME 2 extended instruction sets is mainly the difference in upgraded versions, and they can be applied to existing or new CPU computing acceleration to improve the operational efficiency of AI application services. For example, the execution speed of Whisper-based speech recognition is greatly improved, and the interactive input part of large language models can process more words in the same time, thereby accelerating the speed of AI content generation.
This means that in addition to improving computing performance with new CPU and GPU computing clusters, Arm also emphasizes that it can accelerate artificial intelligence computing efficiency through the SME2 extended instruction set. At the same time, it can also accelerate the execution performance of different computing components through the SME2 extended instruction set, thereby accelerating the development of personalized AI terminal computing.
However, regarding advancements in AI computing, James McNiven emphasized that the key consideration lies in actual application needs and the deployment of appropriate computing components. For example, CPUs, GPUs, and NPUs each have their own strengths in computing. Furthermore, the actual AI application scenarios must be considered, such as accelerating AI computing with a hybrid architecture when connected to the internet, or using device-side computing power to drive AI operations when disconnected from the internet.
Next, we will also support the 2nm process and maintain multiple design licensing methods
Current Lumex computing subsystemCorresponding to 3nm process designAlthough James McNiven did not disclose whether the relevant verification was completed on TSMC or Samsung process technology, he emphasized that he currently maintains in-depth cooperation with many wafer foundries and will even promote support for 2nm process in the future, so that the Lumex computing subsystem can also correspond to more advanced process technology, thereby meeting more processor design requirements.
Regarding current design licensing, James McNiven emphasized that CSS computing subsystem licensing will still be prioritized, but the company will still retain previous RTL licensing and customized licensing to meet the diverse design needs of market players. For example, the company is currently collaborating on various licensing plans with Apple, Samsung, MediaTek, Huawei, and other companies for a long time.
On the other hand, James McNiven also believes that Xiaomi's investment in developing its own processors is "Xuan Ring O1"This approach will become one of the future market development trends, but it is also believed that the current situation where more and more companies want to build their own processors but find it difficult to invest huge resources actually brings many development opportunities to Arm.
No response to market rumors alleging Arm's own chips
Regarding the previous market rumors that Arm is interested inBuild your own processor products, using it as a market reference design to attract more industry players to adopt it, but James McNiven did not respond to this.
