Arm reveals technical details of AGI CPU and Neoverse V3

New technical details have been revealed regarding Arm’s AGI CPU, specifically concerning the internal specifications of the Neoverse V3 core. Based on the Neoverse Compute Subsystems (CSS), the architecture features a 10-wide decode and dispatch capability, placing its scale on par with NVIDIA’s Olympus cores.

The Neoverse V3 pipeline includes a significant number of execution ports, estimated at 19 in total. This includes 8 ALUs, 3 branch units, 4 load/store units, and 4 floating-point units. Notably, the dual 128-bit SIMD configuration allows for SVE2 processing performance of 256-bit per cycle, which exceeds the performance of the Olympus architecture.

Technical observations also highlighted a high L2 bandwidth of 128 bytes per cycle, a figure that has raised questions regarding its necessity relative to the load/store unit width. Additionally, the architecture does not appear to support Simultaneous Multithreading (SMT).