AI Briefing
KO

Arm Unveils C2-Ultra CPU and G2-Ultra NX GPU… Performance Gains Include Clock and Cache Effects

·2026.09.08 11:00

Key point

Arm unveiled the C2-Ultra CPU and G2-Ultra NX GPU, but analysis suggests that a significant portion of the performance improvements stems from increased clock speeds and expanded cache.

Details

Arm unveiled the Neoverse CSS N4 IP for data centers, along with the flagship mobile C2-Ultra CPU and the G2-Ultra NX GPU. The G2-Ultra NX is set to be featured in Xiaomi's XRING O3 chip, scheduled for release on August 24.

C2-Ultra CPU: Limited Architectural Changes

The C2-Ultra claims up to 7% IPC (instructions per clock) improvement, 15% peak performance, and 12% average workload performance gains over its predecessor, the C1-Ultra. However, these figures include the effects of an 8.5% increase in clock speed and an additional 3MB of L2 cache. Excluding the clock increase, the average performance gain is around 3.2%, leading to assessments that the pure microarchitectural improvements are limited.

In terms of architecture, the C2-Ultra maintains the same decoder and execution unit configuration as the Cortex X925, the core from two generations prior, with only iterative improvements to certain structures such as the branch predictor and out-of-order execution buffer. Power efficiency improved by 38%, but this includes benefits from process node and implementation improvements, making the pure architectural contribution unclear.

G2-Ultra NX GPU: Introduction of Matrix Accelerator

The G2-Ultra NX is described by Arm as the 'biggest GPU redesign in seven generations.' While retaining the shader core specifications (128 FMA units), it integrates a matrix accelerator. This can process 1,024 MACs/clock at INT8 precision, and optimizations to the ray tracing units reduced DRAM traffic by 13%. It also introduces Neural Super Sampling (NSS), an ML-based upscaling technology.

In the Xiaomi XRING O3, only half of the shader cores implement the matrix accelerator, which increases core area by approximately 21%. Arm claims 14% gaming performance and 24% ray tracing benchmark improvements over the previous generation, but considering the clock speed is roughly 11% higher, the actual performance gains in non-ray tracing games may be minimal.

Neoverse CSS N4 and Assessment

The server-oriented Neoverse CSS N4 supports 8 to 128 cores per die, with a maximum frequency of 3.8 GHz, DDR5/LPDDR6 memory, PCIe Gen 6/7, and CXL 4.0. However, Arm's announcement materials have been criticized for lacking key technical details, requiring information to be obtained through questions.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.