The ARM has unveiled its next generation CPU and GPU designs and is doing some rebranding at the same time – ARM C1 meets CPU and G1 GPU, which will form ARM Lumax Compute Sub -Social (CSS). “Lumax” is a new branding that will be used for ARM mobile-centric designs (for others, “NIVA” for PC, “Automotive for” Zenna “and so on).
This generation of CPU and GPU design makes a large scale improvement in bzword workload (AI, Ray Tracing), while also provides solid improvement in less attractive workloads.
Note that ARM has also designed an in-house chipset, which we will cover in a separate post.
Let’s first look at the CPU and we will start with a deception sheet to help you keep an eye on the new branding-cortex-X and Cortex-A!
- Cortex-X 9xx → ARM C1-Altra
- (New) ARM C1-Pramium
- Cortex-A7xx → Arm C1-PRO
- Cortex-A5xx → Arm C1-Nano
The new C1 core is first created on Armv9.3 Architecture. They all support the scalable matrix extension 2 (SME 2) of ARM, which accelerates AI workload, but also improves more regular accessories – such as decoding HDR video is 10% more efficient with SME 2.
Arm C1-AULTRA The IPC focuses on increasing instructions. In solo-threaded functions, the C1-Aultra is 25% faster than the Cortex-X925.
The arm also introduces a new tier of the CPU core, ARM C1-PramiumThis is aimed at a sub-flagship designs-it has a 35% small surface area compared to ultra (meaning chips will be cheaper with it), which gives it the best-in-class area efficiency (ie mM of performance to silicon).
Arm C1-Pro The core will be used as a large core for continuous charge in high performance chips and for mid-range designs. Compared to Cortex-A725, it can give 16% more performance in things such as gaming. Alternatively, this video can be 12% more powerful for playback, web browsing and social media.
Arm C1-Nano The changes are almost completely focused on power efficiency-this small core is 26% more powerful than the Cortex-A520. It also improves a minor performance and has a 2% small core area.
The CPU core will be orchestrated by the new C1-Dynamiq shared unit (DSU). It is responsible for things like sharing L3 cash between all core, power management for the core and more. The new DSU enables power saving up to 26% compared to the previous DSU-120.
An Arm C1 CPU cluster can be covered as 1 CPU core and configured as 14 core. Chipset designers can mix and match, combine up to three core types, select between ultra, premium, pro and nano.
The ARM claims that for the real-world workload, a C1 CPU cluster-agni brings up average 15% speed-up on an average of 30% high performance on the industry-agronic benchmark and things like gaming and video streaming. Additionally, it uses an average of 12% less energy for things such as video playback, web browsing and social media compared to the previous generation CPU.
As we have mentioned above, the new C1 core enables improvement in mass performance to AI through SM2 Extension. We are talking up to 4.7x low delays for the whisper base (an automated speech recognition model, ie speech for text), 4.7x for Google’s GEMMA 3 model 4.7x high AI performance and consistency AI for 2.8x fast audio generation (a lesson for audio models that can generate background audio, music and more).
App developers will get better performance on the next-gene hardware “for free” as SM2 support has been integrated into AI Framework, a major AI from Arm, Google, Microsoft, Alibaba and Meta.
According to a recent study, 83% of gamers play on mobile. It is an attractive business, in the context of both sports and hardware. The ARM says that it has sent more than 12 billion GPUs to date till date – and these are the most powerful designed so far.
Arm Mali Ji 1-Altra An introduction of a second generation of Ray Tracing Unit (RTUV2), which doubles the Ray Tresing performance compared to the inside of the immortal-G925 GPU.
Note that Ray is more to present the game scene than tracing, so you can expect to see 40% high frame rate in games using hardware rand. Additionally, RTUV2 is now a separate module and brings multiple power efficiency optimization including a simple – this system can sleep when it becomes inactive.
Ray Tracing on one side, the gardener can give 20% more rasa performance than the G925 in the G1-AULTRA major benchmark. It can also be 9% more powerful. GPU can also be used for AI and thanks to a new GP16 Matrix Compute Path, G1-ALRA AI is 20% faster on Inference.
Here are some more solid numbers on popular sports:
| Arena breakout | +25% |
| Honakai Star Rail | +19% |
| Gainshin effect | +17% |
| Fortnite | +11% |
| Arm in-house game demo (Mori) | +26% |
Arm Mali-G1 GPU can be configured between 1 and 24 Shader Corps. The accuracy of the G1 GPUs Support Arm is Super Resolution (ASR), which is the company’s Temporal Upskaling Technology (Think DLSS). It is already supported by the unrealistic engine 5 and has been integrated into Fortanite.
Of course, you do not yet have a smartphone with C1 CPU and G1 GPU. The ARM says that the next generation hardware will “in consumer equipment in a very near future”.
