The Ultimate Guide to the Arm Neoverse CSS N4 “Falcon” CPU

Arm Neoverse CSS N4 is Arm's newest semi-custom compute subsystem for data center silicon, designed to give cloud and AI companies a validated starting point for building their own processors.

Unlike a finished consumer CPU, CSS N4 is an IP and platform package that partners can configure around their own requirements, including CPU cores, cache, memory, I/O and accelerator connectivity. Arm describes it as its most configurable Neoverse CSS platform yet.

---

What Is Neoverse CSS N4?

Neoverse CSS N4 combines Neoverse N4 CPU cores, interconnect, memory and system IP into a pre-integrated and validated compute subsystem.

The goal is to reduce the engineering work required to develop custom data center silicon. Partners can start with Arm's validated foundation and focus their resources on differentiating their final chip.

CSS N4 is aimed primarily at:

• Cloud computing
• AI infrastructure
• Networking
• DPUs
• Scale-out workloads
• Agentic AI systems

128 Cores Per Die

The biggest headline feature is its scalability.

CSS N4 can be configured from 8 to 128 Neoverse N4 CPU cores per die. This doubles the maximum core count compared with the previous CSS N3 platform.

Arm says N4 can reach frequencies of up to 3.8 GHz, although actual clock speeds depend on the configuration and implementation chosen by the silicon partner.

The platform can also support up to 256 MB of L3 cache per die, depending on configuration.

---

TSMC N3P Manufacturing

CSS N4 is designed for advanced process technology, with current reporting identifying TSMC's N3P node as the manufacturing process for the platform.

Using an advanced node allows partners to target higher compute density and improved efficiency while building increasingly large CPU dies.

The exact finished processor, however, depends on how each Arm partner implements and configures the CSS N4 platform.

Memory and I/O

CSS N4 is designed for high-bandwidth data center systems rather than conventional desktop PCs.

It introduces support for LPDDR6 memory and PCIe Gen 7 connectivity, alongside configurable memory and I/O options.

The platform can also be configured for chiplet connectivity and accelerator attachment, making it suitable for systems where CPUs need to work closely with AI accelerators, networking hardware and other specialized silicon.

Performance and Efficiency

Arm claims significant improvements over Neoverse CSS N3.

For a 128-core configuration running at 3 GHz with 2 MB of L2 cache per core, Arm says CSS N4 can deliver:

These are Arm's claimed improvements over the previous generation, not independent benchmark results.

Why CSS N4 Matters for AI Data Centers

Modern AI infrastructure increasingly needs CPUs to coordinate accelerators, manage data movement, handle networking and run increasingly complex software services.

CSS N4 is designed around that changing role.

Arm specifically positions the platform for agentic AI infrastructure, where CPUs handle orchestration, networking, data management and other workloads around AI accelerators.

Its configurable architecture also allows different companies to create processors optimized for their own infrastructure instead of using an identical off-the-shelf CPU.

CSS N4 vs a Finished CPU

One important distinction is that CSS N4 is not a retail CPU.

Arm provides the compute subsystem as an RTL deliverable along with implementation guidance, IP interoperability support and software resources. A silicon partner then integrates and customizes it as part of its own processor design.

This makes CSS N4 more comparable to a customizable foundation for data center chips than a direct competitor to desktop processors such as Ryzen or Core.

What to Know at a Glance

🎮 The Bottom Line

Arm Neoverse CSS N4 is less about launching another CPU and more about making custom data center CPUs easier to build.

With up to 128 Neoverse N4 cores per die, advanced TSMC N3P manufacturing, LPDDR6, PCIe Gen 7 and extensive configuration options, it gives cloud and AI companies a powerful foundation for their own silicon.

Its real impact will ultimately depend on the processors and systems that Arm's partners build from the platform.